+ + +
Best wishes to you all and thanks so much for your heroic efforts this semester
+ ++
+ +
+ +
+ +
+ + +
+ +
+ +
+ + +
This is an important topic if we aim at extracting a probability +distribution. This gives us also a confidence interval and error +estimates. +
+ +Bayesian machine learning allows us to encode our prior beliefs about +what those models should look like, independent of what the data tells +us. This is especially useful when we don’t have a ton of data to +confidently learn our model. +
+ +Video on Bayesian deep learning + +See also the slides here.
+ ++ +
+ +
+ + +
Reinforcement Learning (RL) is one of the most exciting fields of +Machine Learning today, and also one of the oldest. It has been around +since the 1950s, producing many interesting applications over the +years. +
+ +It studies +how agents take actions based on trial and error, so as to maximize +some notion of cumulative reward in a dynamic system or +environment. Due to its generality, the problem has also been studied +in many other disciplines, such as game theory, control theory, +operations research, information theory, multi-agent systems, swarm +intelligence, statistics, and genetic algorithms. +
+ +In March 2016, AlphaGo, a computer program that plays the board game +Go, beat Lee Sedol in a five-game match. This was the first time a +computer Go program had beaten a 9-dan (highest rank) professional +without handicaps. AlphaGo is based on deep convolutional neural +networks and reinforcement learning. AlphaGo’s victory was a major +milestone in artificial intelligence and it has also made +reinforcement learning a hot research area in the field of machine +learning. +
+ +Lecture on Reinforcement Learning.
+ +See also A. Geron's textbook, chapter 16.
+ ++ +
+ +
+ + +
The goal of transfer learning is to transfer the model or knowledge +obtained from a source task to the target task, in order to resolve +the issues of insufficient training data in the target task. The +rationality of doing so lies in that usually the source and target +tasks have inter-correlations, and therefore either the features, +samples, or models in the source task might provide useful information +for us to better solve the target task. Transfer learning is a hot +research topic in recent years, with many problems still waiting to be studied. +
+ + + ++ +
+ +
+ + +
The conventional deep generative model has a potential problem: the +model tends to generate extreme instances to maximize the +probabilistic likelihood, which will hurt its performance. Adversarial +learning utilizes the adversarial behaviors (e.g., generating +adversarial instances or training an adversarial model) to enhance the +robustness of the model and improve the quality of the generated +data. In recent years, one of the most promising unsupervised learning +technologies, generative adversarial networks (GAN), has already been +successfully applied to image, speech, and text. +
+ +Lecture on adversial learning.
+ ++ +
+ +
+ + +
Dual learning is a new learning paradigm, the basic idea of which is +to use the primal-dual structure between machine learning tasks to +obtain effective feedback/regularization, and guide and strengthen the +learning process, thus reducing the requirement of large-scale labeled +data for deep learning. The idea of dual learning has been applied to +many problems in machine learning, including machine translation, +image style conversion, question answering and generation, image +classification and generation, text classification and generation, +image-to-text, and text-to-image. +
+ ++ +
+ +
+ + +
Distributed computation will speed up machine learning algorithms, +significantly improve their efficiency, and thus enlarge their +application. When distributed meets machine learning, more than just +implementing the machine learning algorithms in parallel is required. +
+ ++ +
+ +
+ + +
Meta learning is an emerging research direction in machine +learning. Roughly speaking, meta learning concerns learning how to +learn, and focuses on the understanding and adaptation of the learning +itself, instead of just completing a specific learning task. That is, +a meta learner needs to be able to evaluate its own learning methods +and adjust its own learning methods according to specific learning +tasks. +
+ ++ +
+ +
+ + +
While there has been much progress in machine learning, there are also challenges.
+ +For example, the mainstream machine learning technologies are +black-box approaches, making us concerned about their potential +risks. To tackle this challenge, we may want to make machine learning +more explainable and controllable. As another example, the +computational complexity of machine learning algorithms is usually +very high and we may want to invent lightweight algorithms or +implementations. Furthermore, in many domains such as physics, +chemistry, biology, and social sciences, people usually seek elegantly +simple equations (e.g., the Schrödinger equation) to uncover the +underlying laws behind various phenomena. In the field of machine +learning, can we reveal simple laws instead of designing more complex +models for data fitting? Although there are many challenges, we are +still very optimistic about the future of machine learning. As we look +forward to the future, here are what we think the research hotspots in +the next ten years will be. +
+ +See the article on Discovery of Physics From Data: Universal Laws and Discrepancies
+ ++ +
+ +
+ + +
Machine learning, especially deep learning, evolves rapidly. The +ability gap between machine and human on many complex cognitive tasks +becomes narrower and narrower. However, we are still in the very early +stage in terms of explaining why those effective models work and how +they work. +
+ +What is missing: the gap between correlation and causation. Standard Machine Learning is based on what e have called a frequentist approach.
+ +Most +machine learning techniques, especially the statistical ones, depend +highly on correlations in data sets to make predictions and analyses. In +contrast, rational humans tend to reply on clear and trustworthy +causality relations obtained via logical reasoning on real and clear +facts. It is one of the core goals of explainable machine learning to +transition from solving problems by data correlation to solving +problems by logical reasoning. +
+ +Bayesian Machine Learning is one of the exciting research directions in this field. + ++ +
+ +
+ + +
An important and emerging field is what has been dubbed as scientific ML, see the article by Deiana et al Applications and Techniques for Fast Machine Learning in Science, arXiv:2110.13041
+ +The authors discuss applications and techniques for fast machine +learning (ML) in science – the concept of integrating power ML +methods into the real-time experimental data processing loop to +accelerate scientific discovery. The report covers three main areas +
+ ++ +
+ +
+ + +
Quantum machine learning is an emerging interdisciplinary research +area at the intersection of quantum computing and machine learning. +
+ +Quantum computers use effects such as quantum coherence and quantum +entanglement to process information, which is fundamentally different +from classical computers. Quantum algorithms have surpassed the best +classical algorithms in several problems (e.g., searching for an +unsorted database, inverting a sparse matrix), which we call quantum +acceleration. +
+ +When quantum computing meets machine learning, it can be a mutually +beneficial and reinforcing process, as it allows us to take advantage +of quantum computing to improve the performance of classical machine +learning algorithms. In addition, we can also use the machine learning +algorithms (on classic computers) to analyze and improve quantum +computing systems. +
+ + + +Read interview with Maria Schuld on her work on Quantum Machine Learning. See also her recent textbook.
+ ++ +
+ +
+ + +
Many quantum machine learning algorithms are based on variants of +quantum algorithms for solving linear equations, which can efficiently +solve N-variable linear equations with complexity of O(log2 N) under +certain conditions. The quantum matrix inversion algorithm can +accelerate many machine learning methods, such as least square linear +regression, least square version of support vector machine, Gaussian +process, and more. The training of these algorithms can be simplified +to solve linear equations. The key bottleneck of this type of quantum +machine learning algorithms is data input—that is, how to initialize +the quantum system with the entire data set. Although efficient +data-input algorithms exist for certain situations, how to efficiently +input data into a quantum system is as yet unknown for most cases. +
+ ++ +
+ +
+ + +
In quantum reinforcement learning, a quantum agent interacts with the +classical environment to obtain rewards from the environment, so as to +adjust and improve its behavioral strategies. In some cases, it +achieves quantum acceleration by the quantum processing capabilities +of the agent or the possibility of exploring the environment through +quantum superposition. Such algorithms have been proposed in +superconducting circuits and systems of trapped ions. +
+ ++ +
+ +
+ + +
Dedicated quantum information processors, such as quantum annealers +and programmable photonic circuits, are well suited for building deep +quantum networks. The simplest deep quantum network is the Boltzmann +machine. The classical Boltzmann machine consists of bits with tunable +interactions and is trained by adjusting the interaction of these bits +so that the distribution of its expression conforms to the statistics +of the data. To quantize the Boltzmann machine, the neural network can +simply be represented as a set of interacting quantum spins that +correspond to an adjustable Ising model. Then, by initializing the +input neurons in the Boltzmann machine to a fixed state and allowing +the system to heat up, we can read out the output qubits to get the +result. +
+ ++ +
+ +
+ + +
Machine learning aims to imitate how humans +learn. While we have developed successful machine learning algorithms, +until now we have ignored one important fact: humans are social. Each +of us is one part of the total society and it is difficult for us to +live, learn, and improve ourselves, alone and isolated. Therefore, we +should design machines with social properties. Can we let machines +evolve by imitating human society so as to achieve more effective, +intelligent, interpretable “social machine learning”? +
+ +And much more.
+ ++ +
+ +
+ + +
Early computer scientist Alan Kay said, The best way to predict the +future is to create it. Therefore, all machine learning +practitioners, whether scholars or engineers, professors or students, +need to work together to advance these important research +topics. Together, we will not just predict the future, but create it. +
+ ++ +
+ +
+ + +
If you wish to have a critical read on AI/ML from a societal point of view, see Kate Crawford's recent text Atlas of AI
+ +Here: with AI/ML we intend a collection of machine learning methods with an emphasis on statistical learning and data analysis + ++ +
+ +
+ + +
+ +
+ +![](\"DataFiles/cancer.png\")
Figure 1:
\n", + "\n", + "\n", + "This tree was produced using the Wisconsin cancer data (discussed here as well, see code examples below) using **Scikit-Learn**'s decision tree classifier. Here we have used the so-called **gini** index (see below) to split the various branches." + ] + }, + { + "cell_type": "markdown", + "id": "f57ac382", + "metadata": {}, + "source": [ + "## General Features\n", + "\n", + "The overarching approach to decision trees is a top-down approach.\n", + "\n", + "* A leaf provides the classification of a given instance.\n", + "\n", + "* A node specifies a test of some attribute of the instance.\n", + "\n", + "* A branch corresponds to a possible values of an attribute.\n", + "\n", + "* An instance is classified by starting at the root node of the tree, testing the attribute specified by this node, then moving down the tree branch corresponding to the value of the attribute in the given example.\n", + "\n", + "This process is then repeated for the subtree rooted at the new\n", + "node." + ] + }, + { + "cell_type": "markdown", + "id": "a9619651", + "metadata": {}, + "source": [ + "## How do we set it up?\n", + "\n", + "In simplified terms, the process of training a decision tree and\n", + "predicting the target features of query instances is as follows:\n", + "\n", + "1. Present a dataset containing of a number of training instances characterized by a number of descriptive features and a target feature\n", + "\n", + "2. Train the decision tree model by continuously splitting the target feature along the values of the descriptive features using a measure of information gain during the training process\n", + "\n", + "3. Grow the tree until we accomplish a stopping criteria create leaf nodes which represent the *predictions* we want to make for new query instances\n", + "\n", + "4. Show query instances to the tree and run down the tree until we arrive at leaf nodes\n", + "\n", + "Then we are essentially done!" + ] + }, + { + "cell_type": "markdown", + "id": "72fccb7b", + "metadata": {}, + "source": [ + "## Decision trees and Regression" + ] + }, + { + "cell_type": "code", + "execution_count": 1, + "id": "53bc6c2a", + "metadata": {}, + "outputs": [], + "source": [ + "%matplotlib inline\n", + "\n", + "import numpy as np\n", + "import matplotlib.pyplot as plt\n", + "from sklearn.preprocessing import PolynomialFeatures\n", + "from sklearn.linear_model import LinearRegression\n", + "\n", + "steps=250\n", + "\n", + "distance=0\n", + "x=0\n", + "distance_list=[]\n", + "steps_list=[]\n", + "while x| Day | Outlook | Temperature | Humidity | Wind | Ride |
|---|---|---|---|---|---|
| 1 | Sunny | Hot | High | Weak | 0 |
| 2 | Sunny | Hot | High | Strong | 1 |
| 3 | Overcast | Hot | High | Weak | 1 |
| 4 | Rain | Mild | High | Weak | 1 |
| 5 | Rain | Cool | Normal | Weak | 1 |
| 6 | Rain | Cool | Normal | Strong | 0 |
| 7 | Overcast | Cool | Normal | Strong | 1 |
| 8 | Sunny | Mild | High | Weak | 0 |
| 9 | Sunny | Cool | Normal | Weak | 1 |
| 10 | Rain | Mild | Normal | Weak | 1 |
| 11 | Sunny | Mild | Normal | Strong | 1 |
| 12 | Overcast | Mild | High | Strong | 1 |
| 13 | Overcast | Hot | Normal | Weak | 1 |
| 14 | Rain | Mild | High | Strong | 0 |
![](\"DataFiles/ensembleoverview.png\")
Figure 1:
\n", + "" + ] + }, + { + "cell_type": "markdown", + "id": "a9001c69", + "metadata": {}, + "source": [ + "## Why Voting?\n", + "\n", + "The idea behind boosting, and voting as well can be phrased as follows:\n", + "**Can a group of people somehow arrive at highly\n", + "reasoned decisions, despite the weak judgement of the individual\n", + "members?**\n", + "\n", + "The aim is to create a good classifier by combining several weak classifiers.\n", + "**A weak classifier is a classifier which is able to produce results that are only slightly better than guessing at random.**\n", + "\n", + "The basic approach is to apply repeatedly (in boosting this is done in an iterative way) a weak classifier to modifications of the data.\n", + "In voting we simply apply the law of large numbers while in boosting we give more weight to misclassified data in\n", + "each iteration. \n", + "\n", + "Decision trees play an important role as our weak classifier. They serve as the basic method." + ] + }, + { + "cell_type": "markdown", + "id": "95bf38e7", + "metadata": {}, + "source": [ + "## Tossing coins\n", + "\n", + "The simplest case is a so-called voting ensemble. To illustrate this,\n", + "think of yourself tossing coins with a biased outcome of 51 per cent\n", + "for heads and 49% for tails. With only few tosses,\n", + "you may not clearly see this distribution for heads and tails. However, after some\n", + "thousands of tosses, there will be a clear majority of heads. With 2000 tosses\n", + "you should see approximately 1020 heads and 980 tails.\n", + "\n", + "We can then state that the outcome is a clear majority of heads. If\n", + "you do this ten thousand times, it is easy to see that there is a 97%\n", + "likelihood of a majority of heads.\n", + "\n", + "Another example would be to collect all polls before an\n", + "election. Different polls may show different likelihoods for a\n", + "candidate winning with say a majority of the popular vote. The majority vote\n", + "would then consist in many polls indicating that this candidate will\n", + "actually win.\n", + "\n", + "The example here shows how we can implement the coin tossing case,\n", + "clealry demostrating that after some tosses we see the [law of large](https://en.wikipedia.org/wiki/Law_of_large_numbers)\n", + "numbers kicking in." + ] + }, + { + "cell_type": "markdown", + "id": "aba2349a", + "metadata": {}, + "source": [ + "## Standard imports first" + ] + }, + { + "cell_type": "code", + "execution_count": 13, + "id": "62312165", + "metadata": {}, + "outputs": [], + "source": [ + "# Common imports\n", + "from IPython.display import Image \n", + "from pydot import graph_from_dot_data\n", + "import pandas as pd\n", + "import numpy as np\n", + "import matplotlib.pyplot as plt\n", + "from sklearn.tree import DecisionTreeClassifier\n", + "from sklearn.model_selection import train_test_split\n", + "from sklearn.tree import export_graphviz\n", + "from sklearn.preprocessing import StandardScaler, OneHotEncoder\n", + "from sklearn.compose import ColumnTransformer\n", + "from IPython.display import Image \n", + "from pydot import graph_from_dot_data\n", + "import os\n", + "\n", + "# Where to save the figures and data files\n", + "PROJECT_ROOT_DIR = \"Results\"\n", + "FIGURE_ID = \"Results/FigureFiles\"\n", + "DATA_ID = \"DataFiles/\"\n", + "\n", + "if not os.path.exists(PROJECT_ROOT_DIR):\n", + " os.mkdir(PROJECT_ROOT_DIR)\n", + "\n", + "if not os.path.exists(FIGURE_ID):\n", + " os.makedirs(FIGURE_ID)\n", + "\n", + "if not os.path.exists(DATA_ID):\n", + " os.makedirs(DATA_ID)\n", + "\n", + "def image_path(fig_id):\n", + " return os.path.join(FIGURE_ID, fig_id)\n", + "\n", + "def data_path(dat_id):\n", + " return os.path.join(DATA_ID, dat_id)\n", + "\n", + "def save_fig(fig_id):\n", + " plt.savefig(image_path(fig_id) + \".png\", format='png')" + ] + }, + { + "cell_type": "markdown", + "id": "7307b530", + "metadata": {}, + "source": [ + "## Simple Voting Example, head or tail" + ] + }, + { + "cell_type": "code", + "execution_count": 14, + "id": "b9bac328", + "metadata": {}, + "outputs": [], + "source": [ + "\n", + "# Common imports\n", + "import numpy as np\n", + "import matplotlib\n", + "import matplotlib.pyplot as plt\n", + "from matplotlib.colors import ListedColormap\n", + "plt.rcParams['axes.labelsize'] = 14\n", + "plt.rcParams['xtick.labelsize'] = 12\n", + "plt.rcParams['ytick.labelsize'] = 12\n", + "\n", + "heads_proba = 0.51\n", + "coin_tosses = (np.random.rand(10000, 10) < heads_proba).astype(np.int32)\n", + "cumulative_heads_ratio = np.cumsum(coin_tosses, axis=0) / np.arange(1, 10001).reshape(-1, 1)\n", + "plt.figure(figsize=(8,3.5))\n", + "plt.plot(cumulative_heads_ratio)\n", + "plt.plot([0, 10000], [0.51, 0.51], \"k--\", linewidth=2, label=\"51%\")\n", + "plt.plot([0, 10000], [0.5, 0.5], \"k-\", label=\"50%\")\n", + "plt.xlabel(\"Number of coin tosses\")\n", + "plt.ylabel(\"Heads ratio\")\n", + "plt.legend(loc=\"lower right\")\n", + "plt.axis([0, 10000, 0.42, 0.58])\n", + "save_fig(\"votingsimple\")\n", + "plt.show()" + ] + }, + { + "cell_type": "markdown", + "id": "8ab7d717", + "metadata": {}, + "source": [ + "## Using the Voting Classifier\n", + "\n", + "We can use the voting classifier on other data sets, here the exciting binary case of two distinct objects using the make moons functionality of **Scikit-Learn**." + ] + }, + { + "cell_type": "code", + "execution_count": 15, + "id": "a4b9c350", + "metadata": {}, + "outputs": [], + "source": [ + "from sklearn.model_selection import train_test_split\n", + "from sklearn.datasets import make_moons\n", + "\n", + "X, y = make_moons(n_samples=500, noise=0.30, random_state=42)\n", + "X_train, X_test, y_train, y_test = train_test_split(X, y, random_state=42)\n", + "\n", + "from sklearn.ensemble import RandomForestClassifier\n", + "from sklearn.ensemble import VotingClassifier\n", + "from sklearn.linear_model import LogisticRegression\n", + "from sklearn.svm import SVC\n", + "\n", + "log_clf = LogisticRegression(solver=\"liblinear\", random_state=42)\n", + "rnd_clf = RandomForestClassifier(n_estimators=10, random_state=42)\n", + "svm_clf = SVC(gamma=\"auto\", random_state=42)\n", + "\n", + "voting_clf = VotingClassifier(\n", + " estimators=[('lr', log_clf), ('rf', rnd_clf), ('svc', svm_clf)],\n", + " voting='hard')\n", + "\n", + "voting_clf.fit(X_train, y_train)\n", + "\n", + "from sklearn.metrics import accuracy_score\n", + "\n", + "for clf in (log_clf, rnd_clf, svm_clf, voting_clf):\n", + " clf.fit(X_train, y_train)\n", + " y_pred = clf.predict(X_test)\n", + " print(clf.__class__.__name__, accuracy_score(y_test, y_pred))\n", + "\n", + "log_clf = LogisticRegression(solver=\"liblinear\", random_state=42)\n", + "rnd_clf = RandomForestClassifier(n_estimators=10, random_state=42)\n", + "svm_clf = SVC(gamma=\"auto\", probability=True, random_state=42)\n", + "voting_clf = VotingClassifier(\n", + " estimators=[('lr', log_clf), ('rf', rnd_clf), ('svc', svm_clf)],\n", + " voting='soft')\n", + "voting_clf.fit(X_train, y_train)\n", + "\n", + "from sklearn.metrics import accuracy_score\n", + "\n", + "for clf in (log_clf, rnd_clf, svm_clf, voting_clf):\n", + " clf.fit(X_train, y_train)\n", + " y_pred = clf.predict(X_test)\n", + " print(clf.__class__.__name__, accuracy_score(y_test, y_pred))" + ] + }, + { + "cell_type": "markdown", + "id": "8b42c140", + "metadata": {}, + "source": [ + "## Voting and Bagging" + ] + }, + { + "cell_type": "code", + "execution_count": 16, + "id": "2d91b7c4", + "metadata": {}, + "outputs": [], + "source": [ + "from sklearn.model_selection import train_test_split\n", + "from sklearn.datasets import make_moons\n", + "\n", + "X, y = make_moons(n_samples=500, noise=0.30, random_state=42)\n", + "X_train, X_test, y_train, y_test = train_test_split(X, y, random_state=42)\n", + "from sklearn.ensemble import RandomForestClassifier\n", + "from sklearn.ensemble import VotingClassifier\n", + "from sklearn.linear_model import LogisticRegression\n", + "from sklearn.svm import SVC\n", + "\n", + "log_clf = LogisticRegression(random_state=42)\n", + "rnd_clf = RandomForestClassifier(random_state=42)\n", + "svm_clf = SVC(random_state=42)\n", + "\n", + "voting_clf = VotingClassifier(\n", + " estimators=[('lr', log_clf), ('rf', rnd_clf), ('svc', svm_clf)],\n", + " voting='hard')\n", + "voting_clf.fit(X_train, y_train)" + ] + }, + { + "cell_type": "code", + "execution_count": 17, + "id": "841ca35e", + "metadata": {}, + "outputs": [], + "source": [ + "from sklearn.metrics import accuracy_score\n", + "\n", + "for clf in (log_clf, rnd_clf, svm_clf, voting_clf):\n", + " clf.fit(X_train, y_train)\n", + " y_pred = clf.predict(X_test)\n", + " print(clf.__class__.__name__, accuracy_score(y_test, y_pred))" + ] + }, + { + "cell_type": "code", + "execution_count": 18, + "id": "f6fb547e", + "metadata": {}, + "outputs": [], + "source": [ + "log_clf = LogisticRegression(random_state=42)\n", + "rnd_clf = RandomForestClassifier(random_state=42)\n", + "svm_clf = SVC(probability=True, random_state=42)\n", + "\n", + "voting_clf = VotingClassifier(\n", + " estimators=[('lr', log_clf), ('rf', rnd_clf), ('svc', svm_clf)],\n", + " voting='soft')\n", + "voting_clf.fit(X_train, y_train)" + ] + }, + { + "cell_type": "code", + "execution_count": 19, + "id": "21d21aef", + "metadata": {}, + "outputs": [], + "source": [ + "from sklearn.metrics import accuracy_score\n", + "\n", + "for clf in (log_clf, rnd_clf, svm_clf, voting_clf):\n", + " clf.fit(X_train, y_train)\n", + " y_pred = clf.predict(X_test)\n", + " print(clf.__class__.__name__, accuracy_score(y_test, y_pred))" + ] + }, + { + "cell_type": "markdown", + "id": "8a08fc9a", + "metadata": {}, + "source": [ + "## Bagging\n", + "\n", + "The **plain** decision trees suffer from high\n", + "variance. This means that if we split the training data into two parts\n", + "at random, and fit a decision tree to both halves, the results that we\n", + "get could be quite different. In contrast, a procedure with low\n", + "variance will yield similar results if applied repeatedly to distinct\n", + "data sets; linear regression tends to have low variance, if the ratio\n", + "of $n$ to $p$ is moderately large. \n", + "\n", + "**Bootstrap aggregation**, or just **bagging**, is a\n", + "general-purpose procedure for reducing the variance of a statistical\n", + "learning method." + ] + }, + { + "cell_type": "markdown", + "id": "189d48af", + "metadata": {}, + "source": [ + "## More bagging\n", + "\n", + "Bagging typically results in improved accuracy\n", + "over prediction using a single tree. Unfortunately, however, it can be\n", + "difficult to interpret the resulting model. Recall that one of the\n", + "advantages of decision trees is the attractive and easily interpreted\n", + "diagram that results.\n", + "\n", + "However, when we bag a large number of trees, it is no longer\n", + "possible to represent the resulting statistical learning procedure\n", + "using a single tree, and it is no longer clear which variables are\n", + "most important to the procedure. Thus, bagging improves prediction\n", + "accuracy at the expense of interpretability. Although the collection\n", + "of bagged trees is much more difficult to interpret than a single\n", + "tree, one can obtain an overall summary of the importance of each\n", + "predictor using the MSE (for bagging regression trees) or the Gini\n", + "index (for bagging classification trees). In the case of bagging\n", + "regression trees, we can record the total amount that the MSE is\n", + "decreased due to splits over a given predictor, averaged over all $B$ possible\n", + "trees. A large value indicates an important predictor. Similarly, in\n", + "the context of bagging classification trees, we can add up the total\n", + "amount that the Gini index is decreased by splits over a given\n", + "predictor, averaged over all $B$ trees." + ] + }, + { + "cell_type": "markdown", + "id": "b6cfe02e", + "metadata": {}, + "source": [ + "## Making your own Bootstrap: Changing the Level of the Decision Tree\n", + "\n", + "Let us bring up our good old boostrap example from the linear regression lectures. We change the linerar regression algorithm with\n", + "a decision tree wth different depths and perform a bootstrap aggregate (in this case we perform as many bootstraps as data points $n$)." + ] + }, + { + "cell_type": "code", + "execution_count": 20, + "id": "609bd17f", + "metadata": {}, + "outputs": [], + "source": [ + "\n", + "import matplotlib.pyplot as plt\n", + "import numpy as np\n", + "from sklearn.model_selection import train_test_split\n", + "from sklearn.pipeline import make_pipeline\n", + "from sklearn.utils import resample\n", + "from sklearn.tree import DecisionTreeRegressor\n", + "\n", + "n = 100\n", + "n_boostraps = 100\n", + "maxdepth = 8\n", + "\n", + "# Make data set.\n", + "x = np.linspace(-3, 3, n).reshape(-1, 1)\n", + "y = np.exp(-x**2) + 1.5 * np.exp(-(x-2)**2)+ np.random.normal(0, 0.1, x.shape)\n", + "error = np.zeros(maxdepth)\n", + "bias = np.zeros(maxdepth)\n", + "variance = np.zeros(maxdepth)\n", + "polydegree = np.zeros(maxdepth)\n", + "X_train, X_test, y_train, y_test = train_test_split(x, y, test_size=0.2)\n", + "\n", + "from sklearn.preprocessing import StandardScaler\n", + "scaler = StandardScaler()\n", + "scaler.fit(X_train)\n", + "X_train_scaled = scaler.transform(X_train)\n", + "X_test_scaled = scaler.transform(X_test)\n", + "\n", + "# we produce a simple tree first as benchmark\n", + "simpletree = DecisionTreeRegressor(max_depth=3) \n", + "simpletree.fit(X_train_scaled, y_train)\n", + "simpleprediction = simpletree.predict(X_test_scaled)\n", + "for degree in range(1,maxdepth):\n", + " model = DecisionTreeRegressor(max_depth=degree) \n", + " y_pred = np.empty((y_test.shape[0], n_boostraps))\n", + " for i in range(n_boostraps):\n", + " x_, y_ = resample(X_train_scaled, y_train)\n", + " model.fit(x_, y_)\n", + " y_pred[:, i] = model.predict(X_test_scaled)#.ravel()\n", + "\n", + " polydegree[degree] = degree\n", + " error[degree] = np.mean( np.mean((y_test - y_pred)**2, axis=1, keepdims=True) )\n", + " bias[degree] = np.mean( (y_test - np.mean(y_pred, axis=1, keepdims=True))**2 )\n", + " variance[degree] = np.mean( np.var(y_pred, axis=1, keepdims=True) )\n", + " print('Polynomial degree:', degree)\n", + " print('Error:', error[degree])\n", + " print('Bias^2:', bias[degree])\n", + " print('Var:', variance[degree])\n", + " print('{} >= {} + {} = {}'.format(error[degree], bias[degree], variance[degree], bias[degree]+variance[degree]))\n", + " \n", + "mse_simpletree= np.mean( np.mean((y_test - simpleprediction)**2))\n", + "print(\"Simple tree:\",mse_simpletree)\n", + "plt.xlim(1,maxdepth)\n", + "plt.plot(polydegree, error, label='MSE')\n", + "plt.plot(polydegree, bias, label='bias')\n", + "plt.plot(polydegree, variance, label='Variance')\n", + "plt.legend()\n", + "save_fig(\"baggingboot\")\n", + "plt.show()" + ] + }, + { + "cell_type": "markdown", + "id": "bd392364", + "metadata": {}, + "source": [ + "## Random forests\n", + "\n", + "Random forests provide an improvement over bagged trees by way of a\n", + "small tweak that decorrelates the trees. \n", + "\n", + "As in bagging, we build a\n", + "number of decision trees on bootstrapped training samples. But when\n", + "building these decision trees, each time a split in a tree is\n", + "considered, a random sample of $m$ predictors is chosen as split\n", + "candidates from the full set of $p$ predictors. The split is allowed to\n", + "use only one of those $m$ predictors. \n", + "\n", + "A fresh sample of $m$ predictors is\n", + "taken at each split, and typically we choose" + ] + }, + { + "cell_type": "markdown", + "id": "5498b4fd", + "metadata": {}, + "source": [ + "$$\n", + "m\\approx \\sqrt{p}.\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "356ce239", + "metadata": {}, + "source": [ + "In building a random forest, at\n", + "each split in the tree, the algorithm is not even allowed to consider\n", + "a majority of the available predictors. \n", + "\n", + "The reason for this is rather clever. Suppose that there is one very\n", + "strong predictor in the data set, along with a number of other\n", + "moderately strong predictors. Then in the collection of bagged\n", + "variable importance random forest trees, most or all of the trees will\n", + "use this strong predictor in the top split. Consequently, all of the\n", + "bagged trees will look quite similar to each other. Hence the\n", + "predictions from the bagged trees will be highly correlated.\n", + "Unfortunately, averaging many highly correlated quantities does not\n", + "lead to as large of a reduction in variance as averaging many\n", + "uncorrelated quantities. In particular, this means that bagging will\n", + "not lead to a substantial reduction in variance over a single tree in\n", + "this setting." + ] + }, + { + "cell_type": "markdown", + "id": "de8ed0de", + "metadata": {}, + "source": [ + "## Random Forest Algorithm\n", + "The algorithm described here can be applied to both classification and regression problems.\n", + "\n", + "We will grow of forest of say $B$ trees.\n", + "1. For $b=1:B$\n", + "\n", + " * Draw a bootstrap sample from the training data organized in our $\\boldsymbol{X}$ matrix.\n", + "\n", + " * We grow then a random forest tree $T_b$ based on the bootstrapped data by repeating the steps outlined till we reach the maximum node size is reached\n", + "\n", + "1. we select $m \\le p$ variables at random from the $p$ predictors/features\n", + "\n", + "2. pick the best split point among the $m$ features using for example the CART algorithm and create a new node\n", + "\n", + "3. split the node into daughter nodes\n", + "\n", + "4. Output then the ensemble of trees $\\{T_b\\}_1^{B}$ and make predictions for either a regression type of problem or a classification type of problem." + ] + }, + { + "cell_type": "markdown", + "id": "b54f39b2", + "metadata": {}, + "source": [ + "## Random Forests Compared with other Methods on the Cancer Data" + ] + }, + { + "cell_type": "code", + "execution_count": 21, + "id": "ebf1724c", + "metadata": {}, + "outputs": [], + "source": [ + "import matplotlib.pyplot as plt\n", + "import numpy as np\n", + "from sklearn.model_selection import train_test_split \n", + "from sklearn.datasets import load_breast_cancer\n", + "from sklearn.svm import SVC\n", + "from sklearn.linear_model import LogisticRegression\n", + "from sklearn.tree import DecisionTreeClassifier\n", + "from sklearn.ensemble import BaggingClassifier\n", + "\n", + "# Load the data\n", + "cancer = load_breast_cancer()\n", + "\n", + "X_train, X_test, y_train, y_test = train_test_split(cancer.data,cancer.target,random_state=0)\n", + "print(X_train.shape)\n", + "print(X_test.shape)\n", + "#define methods\n", + "# Logistic Regression\n", + "logreg = LogisticRegression(solver='lbfgs')\n", + "# Support vector machine\n", + "svm = SVC(gamma='auto', C=100)\n", + "# Decision Trees\n", + "deep_tree_clf = DecisionTreeClassifier(max_depth=None)\n", + "#Scale the data\n", + "from sklearn.preprocessing import StandardScaler\n", + "scaler = StandardScaler()\n", + "scaler.fit(X_train)\n", + "X_train_scaled = scaler.transform(X_train)\n", + "X_test_scaled = scaler.transform(X_test)\n", + "# Logistic Regression\n", + "logreg.fit(X_train_scaled, y_train)\n", + "print(\"Test set accuracy Logistic Regression with scaled data: {:.2f}\".format(logreg.score(X_test_scaled,y_test)))\n", + "# Support Vector Machine\n", + "svm.fit(X_train_scaled, y_train)\n", + "print(\"Test set accuracy SVM with scaled data: {:.2f}\".format(logreg.score(X_test_scaled,y_test)))\n", + "# Decision Trees\n", + "deep_tree_clf.fit(X_train_scaled, y_train)\n", + "print(\"Test set accuracy with Decision Trees and scaled data: {:.2f}\".format(deep_tree_clf.score(X_test_scaled,y_test)))\n", + "\n", + "\n", + "from sklearn.ensemble import RandomForestClassifier\n", + "from sklearn.preprocessing import LabelEncoder\n", + "from sklearn.model_selection import cross_validate\n", + "# Data set not specificied\n", + "#Instantiate the model with 500 trees and entropy as splitting criteria\n", + "Random_Forest_model = RandomForestClassifier(n_estimators=500,criterion=\"entropy\")\n", + "Random_Forest_model.fit(X_train_scaled, y_train)\n", + "#Cross validation\n", + "accuracy = cross_validate(Random_Forest_model,X_test_scaled,y_test,cv=10)['test_score']\n", + "print(accuracy)\n", + "print(\"Test set accuracy with Random Forests and scaled data: {:.2f}\".format(Random_Forest_model.score(X_test_scaled,y_test)))\n", + "\n", + "\n", + "import scikitplot as skplt\n", + "y_pred = Random_Forest_model.predict(X_test_scaled)\n", + "skplt.metrics.plot_confusion_matrix(y_test, y_pred, normalize=True)\n", + "plt.show()\n", + "y_probas = Random_Forest_model.predict_proba(X_test_scaled)\n", + "skplt.metrics.plot_roc(y_test, y_probas)\n", + "plt.show()\n", + "skplt.metrics.plot_cumulative_gain(y_test, y_probas)\n", + "plt.show()" + ] + }, + { + "cell_type": "markdown", + "id": "3c629901", + "metadata": {}, + "source": [ + "Recall that the cumulative gains curve shows the percentage of the\n", + "overall number of cases in a given category *gained* by targeting a\n", + "percentage of the total number of cases.\n", + "\n", + "Similarly, the receiver operating characteristic curve, or ROC curve,\n", + "displays the diagnostic ability of a binary classifier system as its\n", + "discrimination threshold is varied. It plots the true positive rate against the false positive rate." + ] + }, + { + "cell_type": "markdown", + "id": "6717213a", + "metadata": {}, + "source": [ + "## Compare Bagging on Trees with Random Forests" + ] + }, + { + "cell_type": "code", + "execution_count": 22, + "id": "2ee05d8e", + "metadata": {}, + "outputs": [], + "source": [ + "bag_clf = BaggingClassifier(\n", + " DecisionTreeClassifier(splitter=\"random\", max_leaf_nodes=16, random_state=42),\n", + " n_estimators=500, max_samples=1.0, bootstrap=True, n_jobs=-1, random_state=42)" + ] + }, + { + "cell_type": "code", + "execution_count": 23, + "id": "b8c99064", + "metadata": {}, + "outputs": [], + "source": [ + "bag_clf.fit(X_train, y_train)\n", + "y_pred = bag_clf.predict(X_test)\n", + "from sklearn.ensemble import RandomForestClassifier\n", + "rnd_clf = RandomForestClassifier(n_estimators=500, max_leaf_nodes=16, n_jobs=-1, random_state=42)\n", + "rnd_clf.fit(X_train, y_train)\n", + "y_pred_rf = rnd_clf.predict(X_test)\n", + "np.sum(y_pred == y_pred_rf) / len(y_pred)" + ] + }, + { + "cell_type": "markdown", + "id": "99d07558", + "metadata": {}, + "source": [ + "## Boosting, a Bird's Eye View\n", + "\n", + "The basic idea is to combine weak classifiers in order to create a good\n", + "classifier. With a weak classifier we often intend a classifier which\n", + "produces results which are only slightly better than we would get by\n", + "random guesses.\n", + "\n", + "This is done by applying in an iterative way a weak (or a standard\n", + "classifier like decision trees) to modify the data. In each iteration\n", + "we emphasize those observations which are misclassified by weighting\n", + "them with a factor." + ] + }, + { + "cell_type": "markdown", + "id": "4ea2cf75", + "metadata": {}, + "source": [ + "## What is boosting? Additive Modelling/Iterative Fitting\n", + "\n", + "Boosting is a way of fitting an additive expansion in a set of\n", + "elementary basis functions like for example some simple polynomials.\n", + "Assume for example that we have a function" + ] + }, + { + "cell_type": "markdown", + "id": "eacb1356", + "metadata": {}, + "source": [ + "$$\n", + "f_M(x) = \\sum_{i=1}^M \\beta_m b(x;\\gamma_m),\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "e014f7e3", + "metadata": {}, + "source": [ + "where $\\beta_m$ are the expansion parameters to be determined in a\n", + "minimization process and $b(x;\\gamma_m)$ are some simple functions of\n", + "the multivariable parameter $x$ which is characterized by the\n", + "parameters $\\gamma_m$.\n", + "\n", + "As an example, consider the Sigmoid function we used in logistic\n", + "regression. In that case, we can translate the function\n", + "$b(x;\\gamma_m)$ into the Sigmoid function" + ] + }, + { + "cell_type": "markdown", + "id": "d3673f5c", + "metadata": {}, + "source": [ + "$$\n", + "\\sigma(t) = \\frac{1}{1+\\exp{(-t)}},\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "121752e8", + "metadata": {}, + "source": [ + "where $t=\\gamma_0+\\gamma_1 x$ and the parameters $\\gamma_0$ and\n", + "$\\gamma_1$ were determined by the Logistic Regression fitting\n", + "algorithm.\n", + "\n", + "As another example, consider the cost function we defined for linear regression" + ] + }, + { + "cell_type": "markdown", + "id": "7d3c5c69", + "metadata": {}, + "source": [ + "$$\n", + "C(\\boldsymbol{y},\\boldsymbol{f}) = \\frac{1}{n} \\sum_{i=0}^{n-1}(y_i-f(x_i))^2.\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "bd1bc84e", + "metadata": {}, + "source": [ + "In this case the function $f(x)$ was replaced by the design matrix\n", + "$\\boldsymbol{X}$ and the unknown linear regression parameters $\\boldsymbol{\\beta}$,\n", + "that is $\\boldsymbol{f}=\\boldsymbol{X}\\boldsymbol{\\beta}$. In linear regression we can \n", + "simply invert a matrix and obtain the parameters $\\beta$ by" + ] + }, + { + "cell_type": "markdown", + "id": "7135b2c5", + "metadata": {}, + "source": [ + "$$\n", + "\\boldsymbol{\\beta}=\\left(\\boldsymbol{X}^T\\boldsymbol{X}\\right)^{-1}\\boldsymbol{X}^T\\boldsymbol{y}.\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "a3531473", + "metadata": {}, + "source": [ + "In iterative fitting or additive modeling, we minimize the cost function with respect to the parameters $\\beta_m$ and $\\gamma_m$." + ] + }, + { + "cell_type": "markdown", + "id": "4e5e0786", + "metadata": {}, + "source": [ + "## Iterative Fitting, Regression and Squared-error Cost Function\n", + "\n", + "The way we proceed is as follows (here we specialize to the squared-error cost function)\n", + "\n", + "1. Establish a cost function, here $C(\\boldsymbol{y},\\boldsymbol{f}) = \\frac{1}{n} \\sum_{i=0}^{n-1}(y_i-f_M(x_i))^2$ with $f_M(x) = \\sum_{i=1}^M \\beta_m b(x;\\gamma_m)$.\n", + "\n", + "2. Initialize with a guess $f_0(x)$. It could be one or even zero or some random numbers.\n", + "\n", + "3. For $m=1:M$\n", + "\n", + "a. minimize $\\sum_{i=0}^{n-1}(y_i-f_{m-1}(x_i)-\\beta b(x;\\gamma))^2$ wrt $\\gamma$ and $\\beta$\n", + "\n", + "b. This gives the optimal values $\\beta_m$ and $\\gamma_m$\n", + "\n", + "c. Determine then the new values $f_m(x)=f_{m-1}(x) +\\beta_m b(x;\\gamma_m)$\n", + "\n", + "We could use any of the algorithms we have discussed till now. If we\n", + "use trees, $\\gamma$ parameterizes the split variables and split points\n", + "at the internal nodes, and the predictions at the terminal nodes." + ] + }, + { + "cell_type": "markdown", + "id": "bd71f33b", + "metadata": {}, + "source": [ + "## Squared-Error Example and Iterative Fitting\n", + "\n", + "To better understand what happens, let us develop the steps for the iterative fitting using the above squared error function.\n", + "\n", + "For simplicity we assume also that our functions $b(x;\\gamma)=1+\\gamma x$. \n", + "\n", + "This means that for every iteration $m$, we need to optimize" + ] + }, + { + "cell_type": "markdown", + "id": "029244f7", + "metadata": {}, + "source": [ + "$$\n", + "(\\beta_m,\\gamma_m) = \\mathrm{argmin}_{\\beta,\\lambda}\\hspace{0.1cm} \\sum_{i=0}^{n-1}(y_i-f_{m-1}(x_i)-\\beta b(x;\\gamma))^2=\\sum_{i=0}^{n-1}(y_i-f_{m-1}(x_i)-\\beta(1+\\gamma x_i))^2.\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "f72c0d58", + "metadata": {}, + "source": [ + "We start our iteration by simply setting $f_0(x)=0$. \n", + "Taking the derivatives with respect to $\\beta$ and $\\gamma$ we obtain" + ] + }, + { + "cell_type": "markdown", + "id": "1b7d8540", + "metadata": {}, + "source": [ + "$$\n", + "\\frac{\\partial {\\cal C}}{\\partial \\beta} = -2\\sum_{i}(1+\\gamma x_i)(y_i-\\beta(1+\\gamma x_i))=0,\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "2d35f2df", + "metadata": {}, + "source": [ + "and" + ] + }, + { + "cell_type": "markdown", + "id": "b5859a45", + "metadata": {}, + "source": [ + "$$\n", + "\\frac{\\partial {\\cal C}}{\\partial \\gamma} =-2\\sum_{i}\\beta x_i(y_i-\\beta(1+\\gamma x_i))=0.\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "6329b7c2", + "metadata": {}, + "source": [ + "We can then rewrite these equations as (defining $\\boldsymbol{w}=\\boldsymbol{e}+\\gamma \\boldsymbol{x})$ with $\\boldsymbol{e}$ being the unit vector)" + ] + }, + { + "cell_type": "markdown", + "id": "b7cc50c4", + "metadata": {}, + "source": [ + "$$\n", + "\\gamma \\boldsymbol{w}^T(\\boldsymbol{y}-\\beta\\gamma \\boldsymbol{w})=0,\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "2282bbac", + "metadata": {}, + "source": [ + "which gives us $\\beta = \\boldsymbol{w}^T\\boldsymbol{y}/(\\boldsymbol{w}^T\\boldsymbol{w})$. Similarly we have" + ] + }, + { + "cell_type": "markdown", + "id": "917718aa", + "metadata": {}, + "source": [ + "$$\n", + "\\beta\\gamma \\boldsymbol{x}^T(\\boldsymbol{y}-\\beta(1+\\gamma \\boldsymbol{x}))=0,\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "c66d9b04", + "metadata": {}, + "source": [ + "which leads to $\\gamma =(\\boldsymbol{x}^T\\boldsymbol{y}-\\beta\\boldsymbol{x}^T\\boldsymbol{e})/(\\beta\\boldsymbol{x}^T\\boldsymbol{x})$. Inserting\n", + "for $\\beta$ gives us an equation for $\\gamma$. This is a non-linear equation in the unknown $\\gamma$ and has to be solved numerically. \n", + "\n", + "The solution to these two equations gives us in turn $\\beta_1$ and $\\gamma_1$ leading to the new expression for $f_1(x)$ as\n", + "$f_1(x) = \\beta_1(1+\\gamma_1x)$. Doing this $M$ times results in our final estimate for the function $f$." + ] + }, + { + "cell_type": "markdown", + "id": "341567c5", + "metadata": {}, + "source": [ + "## Iterative Fitting, Classification and AdaBoost\n", + "\n", + "Let us consider a binary classification problem with two outcomes $y_i \\in \\{-1,1\\}$ and $i=0,1,2,\\dots,n-1$ as our set of\n", + "observations. We define a classification function $G(x)$ which produces a prediction taking one or the other of the two values \n", + "$\\{-1,1\\}$.\n", + "\n", + "The error rate of the training sample is then" + ] + }, + { + "cell_type": "markdown", + "id": "52afa6c2", + "metadata": {}, + "source": [ + "$$\n", + "\\mathrm{\\overline{err}}=\\frac{1}{n} \\sum_{i=0}^{n-1} I(y_i\\ne G(x_i)).\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "e02b6199", + "metadata": {}, + "source": [ + "The iterative procedure starts with defining a weak classifier whose\n", + "error rate is barely better than random guessing. The iterative\n", + "procedure in boosting is to sequentially apply a weak\n", + "classification algorithm to repeatedly modified versions of the data\n", + "producing a sequence of weak classifiers $G_m(x)$.\n", + "\n", + "Here we will express our function $f(x)$ in terms of $G(x)$. That is" + ] + }, + { + "cell_type": "markdown", + "id": "03761f49", + "metadata": {}, + "source": [ + "$$\n", + "f_M(x) = \\sum_{i=1}^M \\beta_m b(x;\\gamma_m),\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "63757911", + "metadata": {}, + "source": [ + "will be a function of" + ] + }, + { + "cell_type": "markdown", + "id": "4b0c293a", + "metadata": {}, + "source": [ + "$$\n", + "G_M(x) = \\mathrm{sign} \\sum_{i=1}^M \\alpha_m G_m(x).\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "70122b7c", + "metadata": {}, + "source": [ + "## Adaptive Boosting, AdaBoost\n", + "\n", + "In our iterative procedure we define thus" + ] + }, + { + "cell_type": "markdown", + "id": "a9bd3da3", + "metadata": {}, + "source": [ + "$$\n", + "f_m(x) = f_{m-1}(x)+\\beta_mG_m(x).\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "4a67628f", + "metadata": {}, + "source": [ + "The simplest possible cost function which leads (also simple from a computational point of view) to the AdaBoost algorithm is the\n", + "exponential cost/loss function defined as" + ] + }, + { + "cell_type": "markdown", + "id": "ef2cf02f", + "metadata": {}, + "source": [ + "$$\n", + "C(\\boldsymbol{y},\\boldsymbol{f}) = \\sum_{i=0}^{n-1}\\exp{(-y_i(f_{m-1}(x_i)+\\beta G(x_i))}.\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "b5cf283d", + "metadata": {}, + "source": [ + "We optimize $\\beta$ and $G$ for each value of $m=1:M$ as we did in the regression case.\n", + "This is normally done in two steps. Let us however first rewrite the cost function as" + ] + }, + { + "cell_type": "markdown", + "id": "8c326212", + "metadata": {}, + "source": [ + "$$\n", + "C(\\boldsymbol{y},\\boldsymbol{f}) = \\sum_{i=0}^{n-1}w_i^{m}\\exp{(-y_i\\beta G(x_i))},\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "e2e7dbcb", + "metadata": {}, + "source": [ + "where we have defined $w_i^m= \\exp{(-y_if_{m-1}(x_i))}$." + ] + }, + { + "cell_type": "markdown", + "id": "8418bbcb", + "metadata": {}, + "source": [ + "## Building up AdaBoost\n", + "\n", + "First, for any $\\beta > 0$, we optimize $G$ by setting" + ] + }, + { + "cell_type": "markdown", + "id": "f28e7f81", + "metadata": {}, + "source": [ + "$$\n", + "G_m(x) = \\mathrm{sign} \\sum_{i=0}^{n-1} w_i^m I(y_i \\ne G_(x_i)),\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "87059974", + "metadata": {}, + "source": [ + "which is the classifier that minimizes the weighted error rate in predicting $y$.\n", + "\n", + "We can do this by rewriting" + ] + }, + { + "cell_type": "markdown", + "id": "65dc0f70", + "metadata": {}, + "source": [ + "$$\n", + "\\exp{-(\\beta)}\\sum_{y_i=G(x_i)}w_i^m+\\exp{(\\beta)}\\sum_{y_i\\ne G(x_i)}w_i^m,\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "7a409f3d", + "metadata": {}, + "source": [ + "which can be rewritten as" + ] + }, + { + "cell_type": "markdown", + "id": "7b339df7", + "metadata": {}, + "source": [ + "$$\n", + "(\\exp{(\\beta)}-\\exp{-(\\beta)})\\sum_{i=0}^{n-1}w_i^mI(y_i\\ne G(x_i))+\\exp{(-\\beta)}\\sum_{i=0}^{n-1}w_i^m=0,\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "857ad2e1", + "metadata": {}, + "source": [ + "which leads to" + ] + }, + { + "cell_type": "markdown", + "id": "280d3ff0", + "metadata": {}, + "source": [ + "$$\n", + "\\beta_m = \\frac{1}{2}\\log{\\frac{1-\\mathrm{\\overline{err}}}{\\mathrm{\\overline{err}}}},\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "7f08ccc4", + "metadata": {}, + "source": [ + "where we have redefined the error as" + ] + }, + { + "cell_type": "markdown", + "id": "5f3ef2a9", + "metadata": {}, + "source": [ + "$$\n", + "\\mathrm{\\overline{err}}_m=\\frac{1}{n}\\frac{\\sum_{i=0}^{n-1}w_i^mI(y_i\\ne G(x_i)}{\\sum_{i=0}^{n-1}w_i^m},\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "bb658722", + "metadata": {}, + "source": [ + "which leads to an update of" + ] + }, + { + "cell_type": "markdown", + "id": "8fd114af", + "metadata": {}, + "source": [ + "$$\n", + "f_m(x) = f_{m-1}(x) +\\beta_m G_m(x).\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "79579995", + "metadata": {}, + "source": [ + "This leads to the new weights" + ] + }, + { + "cell_type": "markdown", + "id": "b47cdb72", + "metadata": {}, + "source": [ + "$$\n", + "w_i^{m+1} = w_i^m \\exp{(-y_i\\beta_m G_m(x_i))}\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "f8385ba9", + "metadata": {}, + "source": [ + "## Adaptive boosting: AdaBoost, Basic Algorithm\n", + "\n", + "The algorithm here is rather straightforward. Assume that our weak\n", + "classifier is a decision tree and we consider a binary set of outputs\n", + "with $y_i \\in \\{-1,1\\}$ and $i=0,1,2,\\dots,n-1$ as our set of\n", + "observations. Our design matrix is given in terms of the\n", + "feature/predictor vectors\n", + "$\\boldsymbol{X}=[\\boldsymbol{x}_0\\boldsymbol{x}_1\\dots\\boldsymbol{x}_{p-1}]$. Finally, we define also a\n", + "classifier determined by our data via a function $G(x)$. This function tells us how well we are able to classify our outputs/targets $\\boldsymbol{y}$. \n", + "\n", + "We have already defined the misclassification error $\\mathrm{err}$ as" + ] + }, + { + "cell_type": "markdown", + "id": "3b7343ad", + "metadata": {}, + "source": [ + "$$\n", + "\\mathrm{err}=\\frac{1}{n}\\sum_{i=0}^{n-1}I(y_i\\ne G(x_i)),\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "debdf61a", + "metadata": {}, + "source": [ + "where the function $I()$ is one if we misclassify and zero if we classify correctly." + ] + }, + { + "cell_type": "markdown", + "id": "a836f153", + "metadata": {}, + "source": [ + "## Basic Steps of AdaBoost\n", + "\n", + "With the above definitions we are now ready to set up the algorithm for AdaBoost.\n", + "The basic idea is to set up weights which will be used to scale the correctly classified and the misclassified cases.\n", + "1. We start by initializing all weights to $w_i = 1/n$, with $i=0,1,2,\\dots n-1$. It is easy to see that we must have $\\sum_{i=0}^{n-1}w_i = 1$.\n", + "\n", + "2. We rewrite the misclassification error as" + ] + }, + { + "cell_type": "markdown", + "id": "4b5ec620", + "metadata": {}, + "source": [ + "$$\n", + "\\mathrm{\\overline{err}}_m=\\frac{\\sum_{i=0}^{n-1}w_i^m I(y_i\\ne G(x_i))}{\\sum_{i=0}^{n-1}w_i},\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "18f2330a", + "metadata": {}, + "source": [ + "1. Then we start looping over all attempts at classifying, namely we start an iterative process for $m=1:M$, where $M$ is the final number of classifications. Our given classifier could for example be a plain decision tree.\n", + "\n", + "a. Fit then a given classifier to the training set using the weights $w_i$.\n", + "\n", + "b. Compute then $\\mathrm{err}$ and figure out which events are classified properly and which are classified wrongly.\n", + "\n", + "c. Define a quantity $\\alpha_{m} = \\log{(1-\\mathrm{\\overline{err}}_m)/\\mathrm{\\overline{err}}_m}$\n", + "\n", + "d. Set the new weights to $w_i = w_i\\times \\exp{(\\alpha_m I(y_i\\ne G(x_i)}$.\n", + "\n", + "5. Compute the new classifier $G(x)= \\sum_{i=0}^{n-1}\\alpha_m I(y_i\\ne G(x_i)$.\n", + "\n", + "For the iterations with $m \\le 2$ the weights are modified\n", + "individually at each steps. The observations which were misclassified\n", + "at iteration $m-1$ have a weight which is larger than those which were\n", + "classified properly. As this proceeds, the observations which were\n", + "difficult to classifiy correctly are given a larger influence. Each\n", + "new classification step $m$ is then forced to concentrate on those\n", + "observations that are missed in the previous iterations." + ] + }, + { + "cell_type": "markdown", + "id": "34ed9bbb", + "metadata": {}, + "source": [ + "## AdaBoost Examples\n", + "\n", + "Using **Scikit-Learn** it is easy to apply the adaptive boosting algorithm, as done here." + ] + }, + { + "cell_type": "code", + "execution_count": 24, + "id": "923ecc83", + "metadata": {}, + "outputs": [], + "source": [ + "from sklearn.ensemble import AdaBoostClassifier\n", + "\n", + "ada_clf = AdaBoostClassifier(\n", + " DecisionTreeClassifier(max_depth=1), n_estimators=200,\n", + " algorithm=\"SAMME.R\", learning_rate=0.5, random_state=42)\n", + "ada_clf.fit(X_train, y_train)\n", + "\n", + "from sklearn.ensemble import AdaBoostClassifier\n", + "\n", + "ada_clf = AdaBoostClassifier(\n", + " DecisionTreeClassifier(max_depth=1), n_estimators=200,\n", + " algorithm=\"SAMME.R\", learning_rate=0.5, random_state=42)\n", + "ada_clf.fit(X_train_scaled, y_train)\n", + "y_pred = ada_clf.predict(X_test_scaled)\n", + "skplt.metrics.plot_confusion_matrix(y_test, y_pred, normalize=True)\n", + "plt.show()\n", + "y_probas = ada_clf.predict_proba(X_test_scaled)\n", + "skplt.metrics.plot_roc(y_test, y_probas)\n", + "plt.show()\n", + "skplt.metrics.plot_cumulative_gain(y_test, y_probas)\n", + "plt.show()" + ] + } + ], + "metadata": { + "kernelspec": { + "display_name": "Python 3 (ipykernel)", + "language": "python", + "name": "python3" + }, + "language_info": { + "codemirror_mode": { + "name": "ipython", + "version": 3 + }, + "file_extension": ".py", + "mimetype": "text/x-python", + "name": "python", + "nbconvert_exporter": "python", + "pygments_lexer": "ipython3", + "version": "3.9.15" + } + }, + "nbformat": 4, + "nbformat_minor": 5 +} diff --git a/doc/pub/week47/ipynb/DataFiles/bank.csv b/doc/pub/week47/ipynb/DataFiles/bank.csv new file mode 100644 index 000000000..930337395 --- /dev/null +++ b/doc/pub/week47/ipynb/DataFiles/bank.csv @@ -0,0 +1,1372 @@ +3.6216,8.6661,-2.8073,-0.44699,0 +4.5459,8.1674,-2.4586,-1.4621,0 +3.866,-2.6383,1.9242,0.10645,0 +3.4566,9.5228,-4.0112,-3.5944,0 +0.32924,-4.4552,4.5718,-0.9888,0 +4.3684,9.6718,-3.9606,-3.1625,0 +3.5912,3.0129,0.72888,0.56421,0 +2.0922,-6.81,8.4636,-0.60216,0 +3.2032,5.7588,-0.75345,-0.61251,0 +1.5356,9.1772,-2.2718,-0.73535,0 +1.2247,8.7779,-2.2135,-0.80647,0 +3.9899,-2.7066,2.3946,0.86291,0 +1.8993,7.6625,0.15394,-3.1108,0 +-1.5768,10.843,2.5462,-2.9362,0 +3.404,8.7261,-2.9915,-0.57242,0 +4.6765,-3.3895,3.4896,1.4771,0 +2.6719,3.0646,0.37158,0.58619,0 +0.80355,2.8473,4.3439,0.6017,0 +1.4479,-4.8794,8.3428,-2.1086,0 +5.2423,11.0272,-4.353,-4.1013,0 +5.7867,7.8902,-2.6196,-0.48708,0 +0.3292,-4.4552,4.5718,-0.9888,0 +3.9362,10.1622,-3.8235,-4.0172,0 +0.93584,8.8855,-1.6831,-1.6599,0 +4.4338,9.887,-4.6795,-3.7483,0 +0.7057,-5.4981,8.3368,-2.8715,0 +1.1432,-3.7413,5.5777,-0.63578,0 +-0.38214,8.3909,2.1624,-3.7405,0 +6.5633,9.8187,-4.4113,-3.2258,0 +4.8906,-3.3584,3.4202,1.0905,0 +-0.24811,-0.17797,4.9068,0.15429,0 +1.4884,3.6274,3.308,0.48921,0 +4.2969,7.617,-2.3874,-0.96164,0 +-0.96511,9.4111,1.7305,-4.8629,0 +-1.6162,0.80908,8.1628,0.60817,0 +2.4391,6.4417,-0.80743,-0.69139,0 +2.6881,6.0195,-0.46641,-0.69268,0 +3.6289,0.81322,1.6277,0.77627,0 +4.5679,3.1929,-2.1055,0.29653,0 +3.4805,9.7008,-3.7541,-3.4379,0 +4.1711,8.722,-3.0224,-0.59699,0 +-0.2062,9.2207,-3.7044,-6.8103,0 +-0.0068919,9.2931,-0.41243,-1.9638,0 +0.96441,5.8395,2.3235,0.066365,0 +2.8561,6.9176,-0.79372,0.48403,0 +-0.7869,9.5663,-3.7867,-7.5034,0 +2.0843,6.6258,0.48382,-2.2134,0 +-0.7869,9.5663,-3.7867,-7.5034,0 +3.9102,6.065,-2.4534,-0.68234,0 +1.6349,3.286,2.8753,0.087054,0 +4.3239,-4.8835,3.4356,-0.5776,0 +5.262,3.9834,-1.5572,1.0103,0 +3.1452,5.825,-0.51439,-1.4944,0 +2.549,6.1499,-1.1605,-1.2371,0 +4.9264,5.496,-2.4774,-0.50648,0 +4.8265,0.80287,1.6371,1.1875,0 +2.5635,6.7769,-0.61979,0.38576,0 +5.807,5.0097,-2.2384,0.43878,0 +3.1377,-4.1096,4.5701,0.98963,0 +-0.78289,11.3603,-0.37644,-7.0495,0 +2.888,0.44696,4.5907,-0.24398,0 +0.49665,5.527,1.7785,-0.47156,0 +4.2586,11.2962,-4.0943,-4.3457,0 +1.7939,-1.1174,1.5454,-0.26079,0 +5.4021,3.1039,-1.1536,1.5651,0 +2.5367,2.599,2.0938,0.20085,0 +4.6054,-4.0765,2.7587,0.31981,0 +2.4235,9.5332,-3.0789,-2.7746,0 +1.0009,7.7846,-0.28219,-2.6608,0 +0.12326,8.9848,-0.9351,-2.4332,0 +3.9529,-2.3548,2.3792,0.48274,0 +4.1373,0.49248,1.093,1.8276,0 +4.7181,10.0153,-3.9486,-3.8582,0 +4.1654,-3.4495,3.643,1.0879,0 +4.4069,10.9072,-4.5775,-4.4271,0 +2.3066,3.5364,0.57551,0.41938,0 +3.7935,7.9853,-2.5477,-1.872,0 +0.049175,6.1437,1.7828,-0.72113,0 +0.24835,7.6439,0.9885,-0.87371,0 +1.1317,3.9647,3.3979,0.84351,0 +2.8033,9.0862,-3.3668,-1.0224,0 +4.4682,2.2907,0.95766,0.83058,0 +5.0185,8.5978,-2.9375,-1.281,0 +1.8664,7.7763,-0.23849,-2.9634,0 +3.245,6.63,-0.63435,0.86937,0 +4.0296,2.6756,0.80685,0.71679,0 +-1.1313,1.9037,7.5339,1.022,0 +0.87603,6.8141,0.84198,-0.17156,0 +4.1197,-2.7956,2.0707,0.67412,0 +3.8027,0.81529,2.1041,1.0245,0 +1.4806,7.6377,-2.7876,-1.0341,0 +4.0632,3.584,0.72545,0.39481,0 +4.3064,8.2068,-2.7824,-1.4336,0 +2.4486,-6.3175,7.9632,0.20602,0 +3.2718,1.7837,2.1161,0.61334,0 +-0.64472,-4.6062,8.347,-2.7099,0 +2.9543,1.076,0.64577,0.89394,0 +2.1616,-6.8804,8.1517,-0.081048,0 +3.82,10.9279,-4.0112,-5.0284,0 +-2.7419,11.4038,2.5394,-5.5793,0 +3.3669,-5.1856,3.6935,-1.1427,0 +4.5597,-2.4211,2.6413,1.6168,0 +5.1129,-0.49871,0.62863,1.1189,0 +3.3397,-4.6145,3.9823,-0.23751,0 +4.2027,0.22761,0.96108,0.97282,0 +3.5438,1.2395,1.997,2.1547,0 +2.3136,10.6651,-3.5288,-4.7672,0 +-1.8584,7.886,-1.6643,-1.8384,0 +3.106,9.5414,-4.2536,-4.003,0 +2.9163,10.8306,-3.3437,-4.122,0 +3.9922,-4.4676,3.7304,-0.1095,0 +1.518,5.6946,0.094818,-0.026738,0 +3.2351,9.647,-3.2074,-2.5948,0 +4.2188,6.8162,-1.2804,0.76076,0 +1.7819,6.9176,-1.2744,-1.5759,0 +2.5331,2.9135,-0.822,-0.12243,0 +3.8969,7.4163,-1.8245,0.14007,0 +2.108,6.7955,-0.1708,0.4905,0 +2.8969,0.70768,2.29,1.8663,0 +0.9297,-3.7971,4.6429,-0.2957,0 +3.4642,10.6878,-3.4071,-4.109,0 +4.0713,10.4023,-4.1722,-4.7582,0 +-1.4572,9.1214,1.7425,-5.1241,0 +-1.5075,1.9224,7.1466,0.89136,0 +-0.91718,9.9884,1.1804,-5.2263,0 +2.994,7.2011,-1.2153,0.3211,0 +-2.343,12.9516,3.3285,-5.9426,0 +3.7818,-2.8846,2.2558,-0.15734,0 +4.6689,1.3098,0.055404,1.909,0 +3.4663,1.1112,1.7425,1.3388,0 +3.2697,-4.3414,3.6884,-0.29829,0 +5.1302,8.6703,-2.8913,-1.5086,0 +2.0139,6.1416,0.37929,0.56938,0 +0.4339,5.5395,2.033,-0.40432,0 +-1.0401,9.3987,0.85998,-5.3336,0 +4.1605,11.2196,-3.6136,-4.0819,0 +5.438,9.4669,-4.9417,-3.9202,0 +5.032,8.2026,-2.6256,-1.0341,0 +5.2418,10.5388,-4.1174,-4.2797,0 +-0.2062,9.2207,-3.7044,-6.8103,0 +2.0911,0.94358,4.5512,1.234,0 +1.7317,-0.34765,4.1905,-0.99138,0 +4.1736,3.3336,-1.4244,0.60429,0 +3.9232,-3.2467,3.4579,0.83705,0 +3.8481,10.1539,-3.8561,-4.2228,0 +0.5195,-3.2633,3.0895,-0.9849,0 +3.8584,0.78425,1.1033,1.7008,0 +1.7496,-0.1759,5.1827,1.2922,0 +3.6277,0.9829,0.68861,0.63403,0 +2.7391,7.4018,0.071684,-2.5302,0 +4.5447,8.2274,-2.4166,-1.5875,0 +-1.7599,11.9211,2.6756,-3.3241,0 +5.0691,0.21313,0.20278,1.2095,0 +3.4591,11.112,-4.2039,-5.0931,0 +1.9358,8.1654,-0.023425,-2.2586,0 +2.486,-0.99533,5.3404,-0.15475,0 +2.4226,-4.5752,5.947,0.21507,0 +3.9479,-3.7723,2.883,0.019813,0 +2.2634,-4.4862,3.6558,-0.61251,0 +1.3566,4.2358,2.1341,0.3211,0 +5.0452,3.8964,-1.4304,0.86291,0 +3.5499,8.6165,-3.2794,-1.2009,0 +0.17346,7.8695,0.26876,-3.7883,0 +2.4008,9.3593,-3.3565,-3.3526,0 +4.8851,1.5995,-0.00029081,1.6401,0 +4.1927,-3.2674,2.5839,0.21766,0 +1.1166,8.6496,-0.96252,-1.8112,0 +1.0235,6.901,-2.0062,-2.7125,0 +-1.803,11.8818,2.0458,-5.2728,0 +0.11739,6.2761,-1.5495,-2.4746,0 +0.5706,-0.0248,1.2421,-0.5621,0 +4.0552,-2.4583,2.2806,1.0323,0 +-1.6952,1.0657,8.8294,0.94955,0 +-1.1193,10.7271,2.0938,-5.6504,0 +1.8799,2.4707,2.4931,0.37671,0 +3.583,-3.7971,3.4391,-0.12501,0 +0.19081,9.1297,-3.725,-5.8224,0 +3.6582,5.6864,-1.7157,-0.23751,0 +-0.13144,-1.7775,8.3316,0.35214,0 +2.3925,9.798,-3.0361,-2.8224,0 +1.6426,3.0149,0.22849,-0.147,0 +-0.11783,-1.5789,8.03,-0.028031,0 +-0.69572,8.6165,1.8419,-4.3289,0 +2.9421,7.4101,-0.97709,-0.88406,0 +-1.7559,11.9459,3.0946,-4.8978,0 +-1.2537,10.8803,1.931,-4.3237,0 +3.2585,-4.4614,3.8024,-0.15087,0 +1.8314,6.3672,-0.036278,0.049554,0 +4.5645,-3.6275,2.8684,0.27714,0 +2.7365,-5.0325,6.6608,-0.57889,0 +0.9297,-3.7971,4.6429,-0.2957,0 +3.9663,10.1684,-4.1131,-4.6056,0 +1.4578,-0.08485,4.1785,0.59136,0 +4.8272,3.0687,0.68604,0.80731,0 +-2.341,12.3784,0.70403,-7.5836,0 +-1.8584,7.886,-1.6643,-1.8384,0 +4.1454,7.257,-1.9153,-0.86078,0 +1.9157,6.0816,0.23705,-2.0116,0 +4.0215,-2.1914,2.4648,1.1409,0 +5.8862,5.8747,-2.8167,-0.30087,0 +-2.0897,10.8265,2.3603,-3.4198,0 +4.0026,-3.5943,3.5573,0.26809,0 +-0.78689,9.5663,-3.7867,-7.5034,0 +4.1757,10.2615,-3.8552,-4.3056,0 +0.83292,7.5404,0.65005,-0.92544,0 +4.8077,2.2327,-0.26334,1.5534,0 +5.3063,5.2684,-2.8904,-0.52716,0 +2.5605,9.2683,-3.5913,-1.356,0 +2.1059,7.6046,-0.47755,-1.8461,0 +2.1721,-0.73874,5.4672,-0.72371,0 +4.2899,9.1814,-4.6067,-4.3263,0 +3.5156,10.1891,-4.2759,-4.978,0 +2.614,8.0081,-3.7258,-1.3069,0 +0.68087,2.3259,4.9085,0.54998,0 +4.1962,0.74493,0.83256,0.753,0 +6.0919,2.9673,-1.3267,1.4551,0 +1.3234,3.2964,0.2362,-0.11984,0 +1.3264,1.0326,5.6566,-0.41337,0 +-0.16735,7.6274,1.2061,-3.6241,0 +-1.3,10.2678,-2.953,-5.8638,0 +-2.2261,12.5398,2.9438,-3.5258,0 +2.4196,6.4665,-0.75688,0.228,0 +1.0987,0.6394,5.989,-0.58277,0 +4.6464,10.5326,-4.5852,-4.206,0 +-0.36038,4.1158,3.1143,-0.37199,0 +1.3562,3.2136,4.3465,0.78662,0 +0.5706,-0.0248,1.2421,-0.5621,0 +-2.6479,10.1374,-1.331,-5.4707,0 +3.1219,-3.137,1.9259,-0.37458,0 +5.4944,1.5478,0.041694,1.9284,0 +-1.3389,1.552,7.0806,1.031,0 +-2.3361,11.9604,3.0835,-5.4435,0 +2.2596,-0.033118,4.7355,-0.2776,0 +0.46901,-0.63321,7.3848,0.36507,0 +2.7296,2.8701,0.51124,0.5099,0 +2.0466,2.03,2.1761,-0.083634,0 +-1.3274,9.498,2.4408,-5.2689,0 +3.8905,-2.1521,2.6302,1.1047,0 +3.9994,0.90427,1.1693,1.6892,0 +2.3952,9.5083,-3.1783,-3.0086,0 +3.2704,6.9321,-1.0456,0.23447,0 +-1.3931,1.5664,7.5382,0.78403,0 +1.6406,3.5488,1.3964,-0.36424,0 +2.7744,6.8576,-1.0671,0.075416,0 +2.4287,9.3821,-3.2477,-1.4543,0 +4.2134,-2.806,2.0116,0.67412,0 +1.6472,0.48213,4.7449,1.225,0 +2.0597,-0.99326,5.2119,-0.29312,0 +0.3798,0.7098,0.7572,-0.4444,0 +1.0135,8.4551,-1.672,-2.0815,0 +4.5691,-4.4552,3.1769,0.0042961,0 +0.57461,10.1105,-1.6917,-4.3922,0 +0.5734,9.1938,-0.9094,-1.872,0 +5.2868,3.257,-1.3721,1.1668,0 +4.0102,10.6568,-4.1388,-5.0646,0 +4.1425,-3.6792,3.8281,1.6297,0 +3.0934,-2.9177,2.2232,0.22283,0 +2.2034,5.9947,0.53009,0.84998,0 +3.744,0.79459,0.95851,1.0077,0 +3.0329,2.2948,2.1135,0.35084,0 +3.7731,7.2073,-1.6814,-0.94742,0 +3.1557,2.8908,0.59693,0.79825,0 +1.8114,7.6067,-0.9788,-2.4668,0 +4.988,7.2052,-3.2846,-1.1608,0 +2.483,6.6155,-0.79287,-0.90863,0 +1.594,4.7055,1.3758,0.081882,0 +-0.016103,9.7484,0.15394,-1.6134,0 +3.8496,9.7939,-4.1508,-4.4582,0 +0.9297,-3.7971,4.6429,-0.2957,0 +4.9342,2.4107,-0.17594,1.6245,0 +3.8417,10.0215,-4.2699,-4.9159,0 +5.3915,9.9946,-3.8081,-3.3642,0 +4.4072,-0.070365,2.0416,1.1319,0 +2.6946,6.7976,-0.40301,0.44912,0 +5.2756,0.13863,0.12138,1.1435,0 +3.4312,6.2637,-1.9513,-0.36165,0 +4.052,-0.16555,0.45383,0.51248,0 +1.3638,-4.7759,8.4182,-1.8836,0 +0.89566,7.7763,-2.7473,-1.9353,0 +1.9265,7.7557,-0.16823,-3.0771,0 +0.20977,-0.46146,7.7267,0.90946,0 +4.068,-2.9363,2.1992,0.50084,0 +2.877,-4.0599,3.6259,-0.32544,0 +0.3223,-0.89808,8.0883,0.69222,0 +-1.3,10.2678,-2.953,-5.8638,0 +1.7747,-6.4334,8.15,-0.89828,0 +1.3419,-4.4221,8.09,-1.7349,0 +0.89606,10.5471,-1.4175,-4.0327,0 +0.44125,2.9487,4.3225,0.7155,0 +3.2422,6.2265,0.12224,-1.4466,0 +2.5678,3.5136,0.61406,-0.40691,0 +-2.2153,11.9625,0.078538,-7.7853,0 +4.1349,6.1189,-2.4294,-0.19613,0 +1.934,-9.2828e-06,4.816,-0.33967,0 +2.5068,1.1588,3.9249,0.12585,0 +2.1464,6.0795,-0.5778,-2.2302,0 +0.051979,7.0521,-2.0541,-3.1508,0 +1.2706,8.035,-0.19651,-2.1888,0 +1.143,0.83391,5.4552,-0.56984,0 +2.2928,9.0386,-3.2417,-1.2991,0 +0.3292,-4.4552,4.5718,-0.9888,0 +2.9719,6.8369,-0.2702,0.71291,0 +1.6849,8.7489,-1.2641,-1.3858,0 +-1.9177,11.6894,2.5454,-3.2763,0 +2.3729,10.4726,-3.0087,-3.2013,0 +1.0284,9.767,-1.3687,-1.7853,0 +0.27451,9.2186,-3.2863,-4.8448,0 +1.6032,-4.7863,8.5193,-2.1203,0 +4.616,10.1788,-4.2185,-4.4245,0 +4.2478,7.6956,-2.7696,-1.0767,0 +4.0215,-2.7004,2.4957,0.36636,0 +5.0297,-4.9704,3.5025,-0.23751,0 +1.5902,2.2948,3.2403,0.18404,0 +2.1274,5.1939,-1.7971,-1.1763,0 +1.1811,8.3847,-2.0567,-0.90345,0 +0.3292,-4.4552,4.5718,-0.9888,0 +5.7353,5.2808,-2.2598,0.075416,0 +2.6718,5.6574,0.72974,-1.4892,0 +1.5799,-4.7076,7.9186,-1.5487,0 +2.9499,2.2493,1.3458,-0.037083,0 +0.5195,-3.2633,3.0895,-0.9849,0 +3.7352,9.5911,-3.9032,-3.3487,0 +-1.7344,2.0175,7.7618,0.93532,0 +3.884,10.0277,-3.9298,-4.0819,0 +3.5257,1.2829,1.9276,1.7991,0 +4.4549,2.4976,1.0313,0.96894,0 +-0.16108,-6.4624,8.3573,-1.5216,0 +4.2164,9.4607,-4.9288,-5.2366,0 +3.5152,6.8224,-0.67377,-0.46898,0 +1.6988,2.9094,2.9044,0.11033,0 +1.0607,2.4542,2.5188,-0.17027,0 +2.0421,1.2436,4.2171,0.90429,0 +3.5594,1.3078,1.291,1.6556,0 +3.0009,5.8126,-2.2306,-0.66553,0 +3.9294,1.4112,1.8076,0.89782,0 +3.4667,-4.0724,4.2882,1.5418,0 +3.966,3.9213,0.70574,0.33662,0 +1.0191,2.33,4.9334,0.82929,0 +0.96414,5.616,2.2138,-0.12501,0 +1.8205,6.7562,0.0099913,0.39481,0 +4.9923,7.8653,-2.3515,-0.71984,0 +-1.1804,11.5093,0.15565,-6.8194,0 +4.0329,0.23175,0.89082,1.1823,0 +0.66018,10.3878,-1.4029,-3.9151,0 +3.5982,7.1307,-1.3035,0.21248,0 +-1.8584,7.886,-1.6643,-1.8384,0 +4.0972,0.46972,1.6671,0.91593,0 +3.3299,0.91254,1.5806,0.39352,0 +3.1088,3.1122,0.80857,0.4336,0 +-4.2859,8.5234,3.1392,-0.91639,0 +-1.2528,10.2036,2.1787,-5.6038,0 +0.5195,-3.2633,3.0895,-0.9849,0 +0.3292,-4.4552,4.5718,-0.9888,0 +0.88872,5.3449,2.045,-0.19355,0 +3.5458,9.3718,-4.0351,-3.9564,0 +-0.21661,8.0329,1.8848,-3.8853,0 +2.7206,9.0821,-3.3111,-0.96811,0 +3.2051,8.6889,-2.9033,-0.7819,0 +2.6917,10.8161,-3.3,-4.2888,0 +-2.3242,11.5176,1.8231,-5.375,0 +2.7161,-4.2006,4.1914,0.16981,0 +3.3848,3.2674,0.90967,0.25128,0 +1.7452,4.8028,2.0878,0.62627,0 +2.805,0.57732,1.3424,1.2133,0 +5.7823,5.5788,-2.4089,-0.056479,0 +3.8999,1.734,1.6011,0.96765,0 +3.5189,6.332,-1.7791,-0.020273,0 +3.2294,7.7391,-0.37816,-2.5405,0 +3.4985,3.1639,0.22677,-0.1651,0 +2.1948,1.3781,1.1582,0.85774,0 +2.2526,9.9636,-3.1749,-2.9944,0 +4.1529,-3.9358,2.8633,-0.017686,0 +0.74307,11.17,-1.3824,-4.0728,0 +1.9105,8.871,-2.3386,-0.75604,0 +-1.5055,0.070346,6.8681,-0.50648,0 +0.58836,10.7727,-1.3884,-4.3276,0 +3.2303,7.8384,-3.5348,-1.2151,0 +-1.9922,11.6542,2.6542,-5.2107,0 +2.8523,9.0096,-3.761,-3.3371,0 +4.2772,2.4955,0.48554,0.36119,0 +1.5099,0.039307,6.2332,-0.30346,0 +5.4188,10.1457,-4.084,-3.6991,0 +0.86202,2.6963,4.2908,0.54739,0 +3.8117,10.1457,-4.0463,-4.5629,0 +0.54777,10.3754,-1.5435,-4.1633,0 +2.3718,7.4908,0.015989,-1.7414,0 +-2.4953,11.1472,1.9353,-3.4638,0 +4.6361,-2.6611,2.8358,1.1991,0 +-2.2527,11.5321,2.5899,-3.2737,0 +3.7982,10.423,-4.1602,-4.9728,0 +-0.36279,8.2895,-1.9213,-3.3332,0 +2.1265,6.8783,0.44784,-2.2224,0 +0.86736,5.5643,1.6765,-0.16769,0 +3.7831,10.0526,-3.8869,-3.7366,0 +-2.2623,12.1177,0.28846,-7.7581,0 +1.2616,4.4303,-1.3335,-1.7517,0 +2.6799,3.1349,0.34073,0.58489,0 +-0.39816,5.9781,1.3912,-1.1621,0 +4.3937,0.35798,2.0416,1.2004,0 +2.9695,5.6222,0.27561,-1.1556,0 +1.3049,-0.15521,6.4911,-0.75346,0 +2.2123,-5.8395,7.7687,-0.85302,0 +1.9647,6.9383,0.57722,0.66377,0 +3.0864,-2.5845,2.2309,0.30947,0 +0.3798,0.7098,0.7572,-0.4444,0 +0.58982,7.4266,1.2353,-2.9595,0 +0.14783,7.946,1.0742,-3.3409,0 +-0.062025,6.1975,1.099,-1.131,0 +4.223,1.1319,0.72202,0.96118,0 +0.64295,7.1018,0.3493,-0.41337,0 +1.941,0.46351,4.6472,1.0879,0 +4.0047,0.45937,1.3621,1.6181,0 +3.7767,9.7794,-3.9075,-3.5323,0 +3.4769,-0.15314,2.53,2.4495,0 +1.9818,9.2621,-3.521,-1.872,0 +3.8023,-3.8696,4.044,0.95343,0 +4.3483,11.1079,-4.0857,-4.2539,0 +1.1518,1.3864,5.2727,-0.43536,0 +-1.2576,1.5892,7.0078,0.42455,0 +1.9572,-5.1153,8.6127,-1.4297,0 +-2.484,12.1611,2.8204,-3.7418,0 +-1.1497,1.2954,7.701,0.62627,0 +4.8368,10.0132,-4.3239,-4.3276,0 +-0.12196,8.8068,0.94566,-4.2267,0 +1.9429,6.3961,0.092248,0.58102,0 +1.742,-4.809,8.2142,-2.0659,0 +-1.5222,10.8409,2.7827,-4.0974,0 +-1.3,10.2678,-2.953,-5.8638,0 +3.4246,-0.14693,0.80342,0.29136,0 +2.5503,-4.9518,6.3729,-0.41596,0 +1.5691,6.3465,-0.1828,-2.4099,0 +1.3087,4.9228,2.0013,0.22024,0 +5.1776,8.2316,-3.2511,-1.5694,0 +2.229,9.6325,-3.1123,-2.7164,0 +5.6272,10.0857,-4.2931,-3.8142,0 +1.2138,8.7986,-2.1672,-0.74182,0 +0.3798,0.7098,0.7572,-0.4444,0 +0.5415,6.0319,1.6825,-0.46122,0 +4.0524,5.6802,-1.9693,0.026279,0 +4.7285,2.1065,-0.28305,1.5625,0 +3.4359,0.66216,2.1041,1.8922,0 +0.86816,10.2429,-1.4912,-4.0082,0 +3.359,9.8022,-3.8209,-3.7133,0 +3.6702,2.9942,0.85141,0.30688,0 +1.3349,6.1189,0.46497,0.49826,0 +3.1887,-3.4143,2.7742,-0.2026,0 +2.4527,2.9653,0.20021,-0.056479,0 +3.9121,2.9735,0.92852,0.60558,0 +3.9364,10.5885,-3.725,-4.3133,0 +3.9414,-3.2902,3.1674,1.0866,0 +3.6922,-3.9585,4.3439,1.3517,0 +5.681,7.795,-2.6848,-0.92544,0 +0.77124,9.0862,-1.2281,-1.4996,0 +3.5761,9.7753,-3.9795,-3.4638,0 +1.602,6.1251,0.52924,0.47886,0 +2.6682,10.216,-3.4414,-4.0069,0 +2.0007,1.8644,2.6491,0.47369,0 +0.64215,3.1287,4.2933,0.64696,0 +4.3848,-3.0729,3.0423,1.2741,0 +0.77445,9.0552,-2.4089,-1.3884,0 +0.96574,8.393,-1.361,-1.4659,0 +3.0948,8.7324,-2.9007,-0.96682,0 +4.9362,7.6046,-2.3429,-0.85302,0 +-1.9458,11.2217,1.9079,-3.4405,0 +5.7403,-0.44284,0.38015,1.3763,0 +-2.6989,12.1984,0.67661,-8.5482,0 +1.1472,3.5985,1.9387,-0.43406,0 +2.9742,8.96,-2.9024,-1.0379,0 +4.5707,7.2094,-3.2794,-1.4944,0 +0.1848,6.5079,2.0133,-0.87242,0 +0.87256,9.2931,-0.7843,-2.1978,0 +0.39559,6.8866,1.0588,-0.67587,0 +3.8384,6.1851,-2.0439,-0.033204,0 +2.8209,7.3108,-0.81857,-1.8784,0 +2.5817,9.7546,-3.1749,-2.9957,0 +3.8213,0.23175,2.0133,2.0564,0 +0.3798,0.7098,0.7572,-0.4444,0 +3.4893,6.69,-1.2042,-0.38751,0 +-1.7781,0.8546,7.1303,0.027572,0 +2.0962,2.4769,1.9379,-0.040962,0 +0.94732,-0.57113,7.1903,-0.67587,0 +2.8261,9.4007,-3.3034,-1.0509,0 +0.0071249,8.3661,0.50781,-3.8155,0 +0.96788,7.1907,1.2798,-2.4565,0 +4.7432,2.1086,0.1368,1.6543,0 +3.6575,7.2797,-2.2692,-1.144,0 +3.8832,6.4023,-2.432,-0.98363,0 +3.4776,8.811,-3.1886,-0.92285,0 +1.1315,7.9212,1.093,-2.8444,0 +2.8237,2.8597,0.19678,0.57196,0 +1.9321,6.0423,0.26019,-2.053,0 +3.0632,-3.3315,5.1305,0.8267,0 +-1.8411,10.8306,2.769,-3.0901,0 +2.8084,11.3045,-3.3394,-4.4194,0 +2.5698,-4.4076,5.9856,0.078002,0 +-0.12624,10.3216,-3.7121,-6.1185,0 +3.3756,-4.0951,4.367,1.0698,0 +-0.048008,-1.6037,8.4756,0.75558,0 +0.5706,-0.0248,1.2421,-0.5621,0 +0.88444,6.5906,0.55837,-0.44182,0 +3.8644,3.7061,0.70403,0.35214,0 +1.2999,2.5762,2.0107,-0.18967,0 +2.0051,-6.8638,8.132,-0.2401,0 +4.9294,0.27727,0.20792,0.33662,0 +2.8297,6.3485,-0.73546,-0.58665,0 +2.565,8.633,-2.9941,-1.3082,0 +2.093,8.3061,0.022844,-3.2724,0 +4.6014,5.6264,-2.1235,0.19309,0 +5.0617,-0.35799,0.44698,0.99868,0 +-0.2951,9.0489,-0.52725,-2.0789,0 +3.577,2.4004,1.8908,0.73231,0 +3.9433,2.5017,1.5215,0.903,0 +2.6648,10.754,-3.3994,-4.1685,0 +5.9374,6.1664,-2.5905,-0.36553,0 +2.0153,1.8479,3.1375,0.42843,0 +5.8782,5.9409,-2.8544,-0.60863,0 +-2.3983,12.606,2.9464,-5.7888,0 +1.762,4.3682,2.1384,0.75429,0 +4.2406,-2.4852,1.608,0.7155,0 +3.4669,6.87,-1.0568,-0.73147,0 +3.1896,5.7526,-0.18537,-0.30087,0 +0.81356,9.1566,-2.1492,-4.1814,0 +0.52855,0.96427,4.0243,-1.0483,0 +2.1319,-2.0403,2.5574,-0.061652,0 +0.33111,4.5731,2.057,-0.18967,0 +1.2746,8.8172,-1.5323,-1.7957,0 +2.2091,7.4556,-1.3284,-3.3021,0 +2.5328,7.528,-0.41929,-2.6478,0 +3.6244,1.4609,1.3501,1.9284,0 +-1.3885,12.5026,0.69118,-7.5487,0 +5.7227,5.8312,-2.4097,-0.24527,0 +3.3583,10.3567,-3.7301,-3.6991,0 +2.5227,2.2369,2.7236,0.79438,0 +0.045304,6.7334,1.0708,-0.9332,0 +4.8278,7.7598,-2.4491,-1.2216,0 +1.9476,-4.7738,8.527,-1.8668,0 +2.7659,0.66216,4.1494,-0.28406,0 +-0.10648,-0.76771,7.7575,0.64179,0 +0.72252,-0.053811,5.6703,-1.3509,0 +4.2475,1.4816,-0.48355,0.95343,0 +3.9772,0.33521,2.2566,2.1625,0 +3.6667,4.302,0.55923,0.33791,0 +2.8232,10.8513,-3.1466,-3.9784,0 +-1.4217,11.6542,-0.057699,-7.1025,0 +4.2458,1.1981,0.66633,0.94696,0 +4.1038,-4.8069,3.3491,-0.49225,0 +1.4507,8.7903,-2.2324,-0.65259,0 +3.4647,-3.9172,3.9746,0.36119,0 +1.8533,6.1458,1.0176,-2.0401,0 +3.5288,0.71596,1.9507,1.9375,0 +3.9719,1.0367,0.75973,1.0013,0 +3.534,9.3614,-3.6316,-1.2461,0 +3.6894,9.887,-4.0788,-4.3664,0 +3.0672,-4.4117,3.8238,-0.81682,0 +2.6463,-4.8152,6.3549,0.003003,0 +2.2893,3.733,0.6312,-0.39786,0 +1.5673,7.9274,-0.056842,-2.1694,0 +4.0405,0.51524,1.0279,1.106,0 +4.3846,-4.8794,3.3662,-0.029324,0 +2.0165,-0.25246,5.1707,1.0763,0 +4.0446,11.1741,-4.3582,-4.7401,0 +-0.33729,-0.64976,7.6659,0.72326,0 +-2.4604,12.7302,0.91738,-7.6418,0 +4.1195,10.9258,-3.8929,-4.1802,0 +2.0193,0.82356,4.6369,1.4202,0 +1.5701,7.9129,0.29018,-2.1953,0 +2.6415,7.586,-0.28562,-1.6677,0 +5.0214,8.0764,-3.0515,-1.7155,0 +4.3435,3.3295,0.83598,0.64955,0 +1.8238,-6.7748,8.3873,-0.54139,0 +3.9382,0.9291,0.78543,0.6767,0 +2.2517,-5.1422,4.2916,-1.2487,0 +5.504,10.3671,-4.413,-4.0211,0 +2.8521,9.171,-3.6461,-1.2047,0 +1.1676,9.1566,-2.0867,-0.80647,0 +2.6104,8.0081,-0.23592,-1.7608,0 +0.32444,10.067,-1.1982,-4.1284,0 +3.8962,-4.7904,3.3954,-0.53751,0 +2.1752,-0.8091,5.1022,-0.67975,0 +1.1588,8.9331,-2.0807,-1.1272,0 +4.7072,8.2957,-2.5605,-1.4905,0 +-1.9667,11.8052,-0.40472,-7.8719,0 +4.0552,0.40143,1.4563,0.65343,0 +2.3678,-6.839,8.4207,-0.44829,0 +0.33565,6.8369,0.69718,-0.55691,0 +4.3398,-5.3036,3.8803,-0.70432,0 +1.5456,8.5482,0.4187,-2.1784,0 +1.4276,8.3847,-2.0995,-1.9677,0 +-0.27802,8.1881,-3.1338,-2.5276,0 +0.93611,8.6413,-1.6351,-1.3043,0 +4.6352,-3.0087,2.6773,1.212,0 +1.5268,-5.5871,8.6564,-1.722,0 +0.95626,2.4728,4.4578,0.21636,0 +-2.7914,1.7734,6.7756,-0.39915,0 +5.2032,3.5116,-1.2538,1.0129,0 +3.1836,7.2321,-1.0713,-2.5909,0 +0.65497,5.1815,1.0673,-0.42113,0 +5.6084,10.3009,-4.8003,-4.3534,0 +1.105,7.4432,0.41099,-3.0332,0 +3.9292,-2.9156,2.2129,0.30817,0 +1.1558,6.4003,1.5506,0.6961,0 +2.5581,2.6218,1.8513,0.40257,0 +2.7831,10.9796,-3.557,-4.4039,0 +3.7635,2.7811,0.66119,0.34179,0 +-2.6479,10.1374,-1.331,-5.4707,0 +1.0652,8.3682,-1.4004,-1.6509,0 +-1.4275,11.8797,0.41613,-6.9978,0 +5.7456,10.1808,-4.7857,-4.3366,0 +5.086,3.2798,-1.2701,1.1189,0 +3.4092,5.4049,-2.5228,-0.89958,0 +-0.2361,9.3221,2.1307,-4.3793,0 +3.8197,8.9951,-4.383,-4.0327,0 +-1.1391,1.8127,6.9144,0.70127,0 +4.9249,0.68906,0.77344,1.2095,0 +2.5089,6.841,-0.029423,0.44912,0 +-0.2062,9.2207,-3.7044,-6.8103,0 +3.946,6.8514,-1.5443,-0.5582,0 +-0.278,8.1881,-3.1338,-2.5276,0 +1.8592,3.2074,-0.15966,-0.26208,0 +0.56953,7.6294,1.5754,-3.2233,0 +3.4626,-4.449,3.5427,0.15429,0 +3.3951,1.1484,2.1401,2.0862,0 +5.0429,-0.52974,0.50439,1.106,0 +3.7758,7.1783,-1.5195,0.40128,0 +4.6562,7.6398,-2.4243,-1.2384,0 +4.0948,-2.9674,2.3689,0.75429,0 +1.8384,6.063,0.54723,0.51248,0 +2.0153,0.43661,4.5864,-0.3151,0 +3.5251,0.7201,1.6928,0.64438,0 +3.757,-5.4236,3.8255,-1.2526,0 +2.5989,3.5178,0.7623,0.81119,0 +1.8994,0.97462,4.2265,0.81377,0 +3.6941,-3.9482,4.2625,1.1577,0 +4.4295,-2.3507,1.7048,0.90946,0 +6.8248,5.2187,-2.5425,0.5461,0 +1.8967,-2.5163,2.8093,-0.79742,0 +2.1526,-6.1665,8.0831,-0.34355,0 +3.3004,7.0811,-1.3258,0.22283,0 +2.7213,7.05,-0.58808,0.41809,0 +3.8846,-3.0336,2.5334,0.20214,0 +4.1665,-0.4449,0.23448,0.27843,0 +0.94225,5.8561,1.8762,-0.32544,0 +5.1321,-0.031048,0.32616,1.1151,0 +0.38251,6.8121,1.8128,-0.61251,0 +3.0333,-2.5928,2.3183,0.303,0 +2.9233,6.0464,-0.11168,-0.58665,0 +1.162,10.2926,-1.2821,-4.0392,0 +3.7791,2.5762,1.3098,0.5655,0 +0.77765,5.9781,1.1941,-0.3526,0 +-0.38388,-1.0471,8.0514,0.49567,0 +0.21084,9.4359,-0.094543,-1.859,0 +2.9571,-4.5938,5.9068,0.57196,0 +4.6439,-3.3729,2.5976,0.55257,0 +3.3577,-4.3062,6.0241,0.18274,0 +3.5127,2.9073,1.0579,0.40774,0 +2.6562,10.7044,-3.3085,-4.0767,0 +-1.3612,10.694,1.7022,-2.9026,0 +-0.278,8.1881,-3.1338,-2.5276,0 +1.04,-6.9321,8.2888,-1.2991,0 +2.1881,2.7356,1.3278,-0.1832,0 +4.2756,-2.6528,2.1375,0.94437,0 +-0.11996,6.8741,0.91995,-0.6694,0 +2.9736,8.7944,-3.6359,-1.3754,0 +3.7798,-3.3109,2.6491,0.066365,0 +5.3586,3.7557,-1.7345,1.0789,0 +1.8373,6.1292,0.84027,0.55257,0 +1.2262,0.89599,5.7568,-0.11596,0 +-0.048008,-0.56078,7.7215,0.453,0 +0.5706,-0.024841,1.2421,-0.56208,0 +4.3634,0.46351,1.4281,2.0202,0 +3.482,-4.1634,3.5008,-0.078462,0 +0.51947,-3.2633,3.0895,-0.98492,0 +2.3164,-2.628,3.1529,-0.08622,0 +-1.8348,11.0334,3.1863,-4.8888,0 +1.3754,8.8793,-1.9136,-0.53751,0 +-0.16682,5.8974,0.49839,-0.70044,0 +0.29961,7.1328,-0.31475,-1.1828,0 +0.25035,9.3262,-3.6873,-6.2543,0 +2.4673,1.3926,1.7125,0.41421,0 +0.77805,6.6424,-1.1425,-1.0573,0 +3.4465,2.9508,1.0271,0.5461,0 +2.2429,-4.1427,5.2333,-0.40173,0 +3.7321,-3.884,3.3577,-0.0060486,0 +4.3365,-3.584,3.6884,0.74912,0 +-2.0759,10.8223,2.6439,-4.837,0 +4.0715,7.6398,-2.0824,-1.1698,0 +0.76163,5.8209,1.1959,-0.64613,0 +-0.53966,7.3273,0.46583,-1.4543,0 +2.6213,5.7919,0.065686,-1.5759,0 +3.0242,-3.3378,2.5865,-0.54785,0 +5.8519,5.3905,-2.4037,-0.061652,0 +0.5706,-0.0248,1.2421,-0.5621,0 +3.9771,11.1513,-3.9272,-4.3444,0 +1.5478,9.1814,-1.6326,-1.7375,0 +0.74054,0.36625,2.1992,0.48403,0 +0.49571,10.2243,-1.097,-4.0159,0 +1.645,7.8612,-0.87598,-3.5569,0 +3.6077,6.8576,-1.1622,0.28231,0 +3.2403,-3.7082,5.2804,0.41291,0 +3.9166,10.2491,-4.0926,-4.4659,0 +3.9262,6.0299,-2.0156,-0.065531,0 +5.591,10.4643,-4.3839,-4.3379,0 +3.7522,-3.6978,3.9943,1.3051,0 +1.3114,4.5462,2.2935,0.22541,0 +3.7022,6.9942,-1.8511,-0.12889,0 +4.364,-3.1039,2.3757,0.78532,0 +3.5829,1.4423,1.0219,1.4008,0 +4.65,-4.8297,3.4553,-0.25174,0 +5.1731,3.9606,-1.983,0.40774,0 +3.2692,3.4184,0.20706,-0.066824,0 +2.4012,1.6223,3.0312,0.71679,0 +1.7257,-4.4697,8.2219,-1.8073,0 +4.7965,6.9859,-1.9967,-0.35001,0 +4.0962,10.1891,-3.9323,-4.1827,0 +2.5559,3.3605,2.0321,0.26809,0 +3.4916,8.5709,-3.0326,-0.59182,0 +0.5195,-3.2633,3.0895,-0.9849,0 +2.9856,7.2673,-0.409,-2.2431,0 +4.0932,5.4132,-1.8219,0.23576,0 +1.7748,-0.76978,5.5854,1.3039,0 +5.2012,0.32694,0.17965,1.1797,0 +-0.45062,-1.3678,7.0858,-0.40303,0 +4.8451,8.1116,-2.9512,-1.4724,0 +0.74841,7.2756,1.1504,-0.5388,0 +5.1213,8.5565,-3.3917,-1.5474,0 +3.6181,-3.7454,2.8273,-0.71208,0 +0.040498,8.5234,1.4461,-3.9306,0 +-2.6479,10.1374,-1.331,-5.4707,0 +0.37984,0.70975,0.75716,-0.44441,0 +-0.95923,0.091039,6.2204,-1.4828,0 +2.8672,10.0008,-3.2049,-3.1095,0 +1.0182,9.109,-0.62064,-1.7129,0 +-2.7143,11.4535,2.1092,-3.9629,0 +3.8244,-3.1081,2.4537,0.52024,0 +2.7961,2.121,1.8385,0.38317,0 +3.5358,6.7086,-0.81857,0.47886,0 +-0.7056,8.7241,2.2215,-4.5965,0 +4.1542,7.2756,-2.4766,-1.2099,0 +0.92703,9.4318,-0.66263,-1.6728,0 +1.8216,-6.4748,8.0514,-0.41855,0 +-2.4473,12.6247,0.73573,-7.6612,0 +3.5862,-3.0957,2.8093,0.24481,0 +0.66191,9.6594,-0.28819,-1.6638,0 +4.7926,1.7071,-0.051701,1.4926,0 +4.9852,8.3516,-2.5425,-1.2823,0 +0.75736,3.0294,2.9164,-0.068117,0 +4.6499,7.6336,-1.9427,-0.37458,0 +-0.023579,7.1742,0.78457,-0.75734,0 +0.85574,0.0082678,6.6042,-0.53104,0 +0.88298,0.66009,6.0096,-0.43277,0 +4.0422,-4.391,4.7466,1.137,0 +2.2546,8.0992,-0.24877,-3.2698,0 +0.38478,6.5989,-0.3336,-0.56466,0 +3.1541,-5.1711,6.5991,0.57455,0 +2.3969,0.23589,4.8477,1.437,0 +4.7114,2.0755,-0.2702,1.2379,0 +4.0127,10.1477,-3.9366,-4.0728,0 +2.6606,3.1681,1.9619,0.18662,0 +3.931,1.8541,-0.023425,1.2314,0 +0.01727,8.693,1.3989,-3.9668,0 +3.2414,0.40971,1.4015,1.1952,0 +2.2504,3.5757,0.35273,0.2836,0 +-1.3971,3.3191,-1.3927,-1.9948,1 +0.39012,-0.14279,-0.031994,0.35084,1 +-1.6677,-7.1535,7.8929,0.96765,1 +-3.8483,-12.8047,15.6824,-1.281,1 +-3.5681,-8.213,10.083,0.96765,1 +-2.2804,-0.30626,1.3347,1.3763,1 +-1.7582,2.7397,-2.5323,-2.234,1 +-0.89409,3.1991,-1.8219,-2.9452,1 +0.3434,0.12415,-0.28733,0.14654,1 +-0.9854,-6.661,5.8245,0.5461,1 +-2.4115,-9.1359,9.3444,-0.65259,1 +-1.5252,-6.2534,5.3524,0.59912,1 +-0.61442,-0.091058,-0.31818,0.50214,1 +-0.36506,2.8928,-3.6461,-3.0603,1 +-5.9034,6.5679,0.67661,-6.6797,1 +-1.8215,2.7521,-0.72261,-2.353,1 +-0.77461,-1.8768,2.4023,1.1319,1 +-1.8187,-9.0366,9.0162,-0.12243,1 +-3.5801,-12.9309,13.1779,-2.5677,1 +-1.8219,-6.8824,5.4681,0.057313,1 +-0.3481,-0.38696,-0.47841,0.62627,1 +0.47368,3.3605,-4.5064,-4.0431,1 +-3.4083,4.8587,-0.76888,-4.8668,1 +-1.6662,-0.30005,1.4238,0.024986,1 +-2.0962,-7.1059,6.6188,-0.33708,1 +-2.6685,-10.4519,9.1139,-1.7323,1 +-0.47465,-4.3496,1.9901,0.7517,1 +1.0552,1.1857,-2.6411,0.11033,1 +1.1644,3.8095,-4.9408,-4.0909,1 +-4.4779,7.3708,-0.31218,-6.7754,1 +-2.7338,0.45523,2.4391,0.21766,1 +-2.286,-5.4484,5.8039,0.88231,1 +-1.6244,-6.3444,4.6575,0.16981,1 +0.50813,0.47799,-1.9804,0.57714,1 +1.6408,4.2503,-4.9023,-2.6621,1 +0.81583,4.84,-5.2613,-6.0823,1 +-5.4901,9.1048,-0.38758,-5.9763,1 +-3.2238,2.7935,0.32274,-0.86078,1 +-2.0631,-1.5147,1.219,0.44524,1 +-0.91318,-2.0113,-0.19565,0.066365,1 +0.6005,1.9327,-3.2888,-0.32415,1 +0.91315,3.3377,-4.0557,-1.6741,1 +-0.28015,3.0729,-3.3857,-2.9155,1 +-3.6085,3.3253,-0.51954,-3.5737,1 +-6.2003,8.6806,0.0091344,-3.703,1 +-4.2932,3.3419,0.77258,-0.99785,1 +-3.0265,-0.062088,0.68604,-0.055186,1 +-1.7015,-0.010356,-0.99337,-0.53104,1 +-0.64326,2.4748,-2.9452,-1.0276,1 +-0.86339,1.9348,-2.3729,-1.0897,1 +-2.0659,1.0512,-0.46298,-1.0974,1 +-2.1333,1.5685,-0.084261,-1.7453,1 +-1.2568,-1.4733,2.8718,0.44653,1 +-3.1128,-6.841,10.7402,-1.0172,1 +-4.8554,-5.9037,10.9818,-0.82199,1 +-2.588,3.8654,-0.3336,-1.2797,1 +0.24394,1.4733,-1.4192,-0.58535,1 +-1.5322,-5.0966,6.6779,0.17498,1 +-4.0025,-13.4979,17.6772,-3.3202,1 +-4.0173,-8.3123,12.4547,-1.4375,1 +-3.0731,-0.53181,2.3877,0.77627,1 +-1.979,3.2301,-1.3575,-2.5819,1 +-0.4294,-0.14693,0.044265,-0.15605,1 +-2.234,-7.0314,7.4936,0.61334,1 +-4.211,-12.4736,14.9704,-1.3884,1 +-3.8073,-8.0971,10.1772,0.65084,1 +-2.5912,-0.10554,1.2798,1.0414,1 +-2.2482,3.0915,-2.3969,-2.6711,1 +-1.4427,3.2922,-1.9702,-3.4392,1 +-0.39416,-0.020702,-0.066267,-0.44699,1 +-1.522,-6.6383,5.7491,-0.10691,1 +-2.8267,-9.0407,9.0694,-0.98233,1 +-1.7263,-6.0237,5.2419,0.29524,1 +-0.94255,0.039307,-0.24192,0.31593,1 +-0.89569,3.0025,-3.6067,-3.4457,1 +-6.2815,6.6651,0.52581,-7.0107,1 +-2.3211,3.166,-1.0002,-2.7151,1 +-1.3414,-2.0776,2.8093,0.60688,1 +-2.258,-9.3263,9.3727,-0.85949,1 +-3.8858,-12.8461,12.7957,-3.1353,1 +-1.8969,-6.7893,5.2761,-0.32544,1 +-0.52645,-0.24832,-0.45613,0.41938,1 +0.0096613,3.5612,-4.407,-4.4103,1 +-3.8826,4.898,-0.92311,-5.0801,1 +-2.1405,-0.16762,1.321,-0.20906,1 +-2.4824,-7.3046,6.839,-0.59053,1 +-2.9098,-10.0712,8.4156,-1.9948,1 +-0.60975,-4.002,1.8471,0.6017,1 +0.83625,1.1071,-2.4706,-0.062945,1 +0.60731,3.9544,-4.772,-4.4853,1 +-4.8861,7.0542,-0.17252,-6.959,1 +-3.1366,0.42212,2.6225,-0.064238,1 +-2.5754,-5.6574,6.103,0.65214,1 +-1.8782,-6.5865,4.8486,-0.021566,1 +0.24261,0.57318,-1.9402,0.44007,1 +1.296,4.2855,-4.8457,-2.9013,1 +0.25943,5.0097,-5.0394,-6.3862,1 +-5.873,9.1752,-0.27448,-6.0422,1 +-3.4605,2.6901,0.16165,-1.0224,1 +-2.3797,-1.4402,1.1273,0.16076,1 +-1.2424,-1.7175,-0.52553,-0.21036,1 +0.20216,1.9182,-3.2828,-0.61768,1 +0.59823,3.5012,-3.9795,-1.7841,1 +-0.77995,3.2322,-3.282,-3.1004,1 +-4.1409,3.4619,-0.47841,-3.8879,1 +-6.5084,8.7696,0.23191,-3.937,1 +-4.4996,3.4288,0.56265,-1.1672,1 +-3.3125,0.10139,0.55323,-0.2957,1 +-1.9423,0.3766,-1.2898,-0.82458,1 +-0.75793,2.5349,-3.0464,-1.2629,1 +-0.95403,1.9824,-2.3163,-1.1957,1 +-2.2173,1.4671,-0.72689,-1.1724,1 +-2.799,1.9679,-0.42357,-2.1125,1 +-1.8629,-0.84841,2.5377,0.097399,1 +-3.5916,-6.2285,10.2389,-1.1543,1 +-5.1216,-5.3118,10.3846,-1.0612,1 +-3.2854,4.0372,-0.45356,-1.8228,1 +-0.56877,1.4174,-1.4252,-1.1246,1 +-2.3518,-4.8359,6.6479,-0.060358,1 +-4.4861,-13.2889,17.3087,-3.2194,1 +-4.3876,-7.7267,11.9655,-1.4543,1 +-3.3604,-0.32696,2.1324,0.6017,1 +-1.0112,2.9984,-1.1664,-1.6185,1 +0.030219,-1.0512,1.4024,0.77369,1 +-1.6514,-8.4985,9.1122,1.2379,1 +-3.2692,-12.7406,15.5573,-0.14182,1 +-2.5701,-6.8452,8.9999,2.1353,1 +-1.3066,0.25244,0.7623,1.7758,1 +-1.6637,3.2881,-2.2701,-2.2224,1 +-0.55008,2.8659,-1.6488,-2.4319,1 +0.21431,-0.69529,0.87711,0.29653,1 +-0.77288,-7.4473,6.492,0.36119,1 +-1.8391,-9.0883,9.2416,-0.10432,1 +-0.63298,-5.1277,4.5624,1.4797,1 +0.0040545,0.62905,-0.64121,0.75817,1 +-0.28696,3.1784,-3.5767,-3.1896,1 +-5.2406,6.6258,-0.19908,-6.8607,1 +-1.4446,2.1438,-0.47241,-1.6677,1 +-0.65767,-2.8018,3.7115,0.99739,1 +-1.5449,-10.1498,9.6152,-1.2332,1 +-2.8957,-12.0205,11.9149,-2.7552,1 +-0.81479,-5.7381,4.3919,0.3211,1 +0.50225,0.65388,-1.1793,0.39998,1 +0.74521,3.6357,-4.4044,-4.1414,1 +-2.9146,4.0537,-0.45699,-4.0327,1 +-1.3907,-1.3781,2.3055,-0.021566,1 +-1.786,-8.1157,7.0858,-1.2112,1 +-1.7322,-9.2828,7.719,-1.7168,1 +0.55298,-3.4619,1.7048,1.1008,1 +2.031,1.852,-3.0121,0.003003,1 +1.2279,4.0309,-4.6435,-3.9125,1 +-4.2249,6.2699,0.15822,-5.5457,1 +-2.5346,-0.77392,3.3602,0.00171,1 +-1.749,-6.332,6.0987,0.14266,1 +-0.539,-5.167,3.4399,0.052141,1 +1.5631,0.89599,-1.9702,0.65472,1 +2.3917,4.5565,-4.9888,-2.8987,1 +0.89512,4.7738,-4.8431,-5.5909,1 +-5.4808,8.1819,0.27818,-5.0323,1 +-2.8833,1.7713,0.68946,-0.4638,1 +-1.4174,-2.2535,1.518,0.61981,1 +0.4283,-0.94981,-1.0731,0.3211,1 +1.5904,2.2121,-3.1183,-0.11725,1 +1.7425,3.6833,-4.0129,-1.7207,1 +-0.23356,3.2405,-3.0669,-2.7784,1 +-3.6227,3.9958,-0.35845,-3.9047,1 +-6.1536,7.9295,0.61663,-3.2646,1 +-3.9172,2.6652,0.78886,-0.7819,1 +-2.2214,-0.23798,0.56008,0.05602,1 +-0.49241,0.89392,-1.6283,-0.56854,1 +0.26517,2.4066,-2.8416,-0.59958,1 +-0.10234,1.8189,-2.2169,-0.56725,1 +-1.6176,1.0926,-0.35502,-0.59958,1 +-1.8448,1.254,0.27218,-1.0728,1 +-1.2786,-2.4087,4.5735,0.47627,1 +-2.902,-7.6563,11.8318,-0.84268,1 +-4.3773,-5.5167,10.939,-0.4082,1 +-2.0529,3.8385,-0.79544,-1.2138,1 +0.18868,0.70148,-0.51182,0.0055892,1 +-1.7279,-6.841,8.9494,0.68058,1 +-3.3793,-13.7731,17.9274,-2.0323,1 +-3.1273,-7.1121,11.3897,-0.083634,1 +-2.121,-0.05588,1.949,1.353,1 +-1.7697,3.4329,-1.2144,-2.3789,1 +-0.0012852,0.13863,-0.19651,0.0081754,1 +-1.682,-6.8121,7.1398,1.3323,1 +-3.4917,-12.1736,14.3689,-0.61639,1 +-3.1158,-8.6289,10.4403,0.97153,1 +-2.0891,-0.48422,1.704,1.7435,1 +-1.6936,2.7852,-2.1835,-1.9276,1 +-1.2846,3.2715,-1.7671,-3.2608,1 +-0.092194,0.39315,-0.32846,-0.13794,1 +-1.0292,-6.3879,5.5255,0.79955,1 +-2.2083,-9.1069,8.9991,-0.28406,1 +-1.0744,-6.3113,5.355,0.80472,1 +-0.51003,-0.23591,0.020273,0.76334,1 +-0.36372,3.0439,-3.4816,-2.7836,1 +-6.3979,6.4479,1.0836,-6.6176,1 +-2.2501,3.3129,-0.88369,-2.8974,1 +-1.1859,-1.2519,2.2635,0.77239,1 +-1.8076,-8.8131,8.7086,-0.21682,1 +-3.3863,-12.9889,13.0545,-2.7202,1 +-1.4106,-7.108,5.6454,0.31335,1 +-0.21394,-0.68287,0.096532,1.1965,1 +0.48797,3.5674,-4.3882,-3.8116,1 +-3.8167,5.1401,-0.65063,-5.4306,1 +-1.9555,0.20692,1.2473,-0.3707,1 +-2.1786,-6.4479,6.0344,-0.20777,1 +-2.3299,-9.9532,8.4756,-1.8733,1 +0.0031201,-4.0061,1.7956,0.91722,1 +1.3518,1.0595,-2.3437,0.39998,1 +1.2309,3.8923,-4.8277,-4.0069,1 +-5.0301,7.5032,-0.13396,-7.5034,1 +-3.0799,0.60836,2.7039,-0.23751,1 +-2.2987,-5.227,5.63,0.91722,1 +-1.239,-6.541,4.8151,-0.033204,1 +0.75896,0.29176,-1.6506,0.83834,1 +1.6799,4.2068,-4.5398,-2.3931,1 +0.63655,5.2022,-5.2159,-6.1211,1 +-6.0598,9.2952,-0.43642,-6.3694,1 +-3.518,2.8763,0.1548,-1.2086,1 +-2.0336,-1.4092,1.1582,0.36507,1 +-0.69745,-1.7672,-0.34474,-0.12372,1 +0.75108,1.9161,-3.1098,-0.20518,1 +0.84546,3.4826,-3.6307,-1.3961,1 +-0.55648,3.2136,-3.3085,-2.7965,1 +-3.6817,3.2239,-0.69347,-3.4004,1 +-6.7526,8.8172,-0.061983,-3.725,1 +-4.577,3.4515,0.66719,-0.94742,1 +-2.9883,0.31245,0.45041,0.068951,1 +-1.4781,0.14277,-1.1622,-0.48579,1 +-0.46651,2.3383,-2.9812,-1.0431,1 +-0.8734,1.6533,-2.1964,-0.78061,1 +-2.1234,1.1815,-0.55552,-0.81165,1 +-2.3142,2.0838,-0.46813,-1.6767,1 +-1.4233,-0.98912,2.3586,0.39481,1 +-3.0866,-6.6362,10.5405,-0.89182,1 +-4.7331,-6.1789,11.388,-1.0741,1 +-2.8829,3.8964,-0.1888,-1.1672,1 +-0.036127,1.525,-1.4089,-0.76121,1 +-1.7104,-4.778,6.2109,0.3974,1 +-3.8203,-13.0551,16.9583,-2.3052,1 +-3.7181,-8.5089,12.363,-0.95518,1 +-2.899,-0.60424,2.6045,1.3776,1 +-0.98193,2.7956,-1.2341,-1.5668,1 +-0.17296,-1.1816,1.3818,0.7336,1 +-1.9409,-8.6848,9.155,0.94049,1 +-3.5713,-12.4922,14.8881,-0.47027,1 +-2.9915,-6.6258,8.6521,1.8198,1 +-1.8483,0.31038,0.77344,1.4189,1 +-2.2677,3.2964,-2.2563,-2.4642,1 +-0.50816,2.868,-1.8108,-2.2612,1 +0.14329,-1.0885,1.0039,0.48791,1 +-0.90784,-7.9026,6.7807,0.34179,1 +-2.0042,-9.3676,9.3333,-0.10303,1 +-0.93587,-5.1008,4.5367,1.3866,1 +-0.40804,0.54214,-0.52725,0.6586,1 +-0.8172,3.3812,-3.6684,-3.456,1 +-4.8392,6.6755,-0.24278,-6.5775,1 +-1.2792,2.1376,-0.47584,-1.3974,1 +-0.66008,-3.226,3.8058,1.1836,1 +-1.7713,-10.7665,10.2184,-1.0043,1 +-3.0061,-12.2377,11.9552,-2.1603,1 +-1.1022,-5.8395,4.5641,0.68705,1 +0.11806,0.39108,-0.98223,0.42843,1 +0.11686,3.735,-4.4379,-4.3741,1 +-2.7264,3.9213,-0.49212,-3.6371,1 +-1.2369,-1.6906,2.518,0.51636,1 +-1.8439,-8.6475,7.6796,-0.66682,1 +-1.8554,-9.6035,7.7764,-0.97716,1 +0.16358,-3.3584,1.3749,1.3569,1 +1.5077,1.9596,-3.0584,-0.12243,1 +0.67886,4.1199,-4.569,-4.1414,1 +-3.9934,5.8333,0.54723,-4.9379,1 +-2.3898,-0.78427,3.0141,0.76205,1 +-1.7976,-6.7686,6.6753,0.89912,1 +-0.70867,-5.5602,4.0483,0.903,1 +1.0194,1.1029,-2.3,0.59395,1 +1.7875,4.78,-5.1362,-3.2362,1 +0.27331,4.8773,-4.9194,-5.8198,1 +-5.1661,8.0433,0.044265,-4.4983,1 +-2.7028,1.6327,0.83598,-0.091393,1 +-1.4904,-2.2183,1.6054,0.89394,1 +-0.014902,-1.0243,-0.94024,0.64955,1 +0.88992,2.2638,-3.1046,-0.11855,1 +1.0637,3.6957,-4.1594,-1.9379,1 +-0.8471,3.1329,-3.0112,-2.9388,1 +-3.9594,4.0289,-0.35845,-3.8957,1 +-5.8818,7.6584,0.5558,-2.9155,1 +-3.7747,2.5162,0.83341,-0.30993,1 +-2.4198,-0.24418,0.70146,0.41809,1 +-0.83535,0.80494,-1.6411,-0.19225,1 +-0.30432,2.6528,-2.7756,-0.65647,1 +-0.60254,1.7237,-2.1501,-0.77027,1 +-2.1059,1.1815,-0.53324,-0.82716,1 +-2.0441,1.2271,0.18564,-1.091,1 +-1.5621,-2.2121,4.2591,0.27972,1 +-3.2305,-7.2135,11.6433,-0.94613,1 +-4.8426,-4.9932,10.4052,-0.53104,1 +-2.3147,3.6668,-0.6969,-1.2474,1 +-0.11716,0.60422,-0.38587,-0.059065,1 +-2.0066,-6.719,9.0162,0.099985,1 +-3.6961,-13.6779,17.5795,-2.6181,1 +-3.6012,-6.5389,10.5234,-0.48967,1 +-2.6286,0.18002,1.7956,0.97282,1 +-0.82601,2.9611,-1.2864,-1.4647,1 +0.31803,-0.99326,1.0947,0.88619,1 +-1.4454,-8.4385,8.8483,0.96894,1 +-3.1423,-13.0365,15.6773,-0.66165,1 +-2.5373,-6.959,8.8054,1.5289,1 +-1.366,0.18416,0.90539,1.5806,1 +-1.7064,3.3088,-2.2829,-2.1978,1 +-0.41965,2.9094,-1.7859,-2.2069,1 +0.37637,-0.82358,0.78543,0.74524,1 +-0.55355,-7.9233,6.7156,0.74394,1 +-1.6001,-9.5828,9.4044,0.081882,1 +-0.37013,-5.554,4.7749,1.547,1 +0.12126,0.22347,-0.47327,0.97024,1 +-0.27068,3.2674,-3.5562,-3.0888,1 +-5.119,6.6486,-0.049987,-6.5206,1 +-1.3946,2.3134,-0.44499,-1.4905,1 +-0.69879,-3.3771,4.1211,1.5043,1 +-1.48,-10.5244,9.9176,-0.5026,1 +-2.6649,-12.813,12.6689,-1.9082,1 +-0.62684,-6.301,4.7843,1.106,1 +0.518,0.25865,-0.84085,0.96118,1 +0.64376,3.764,-4.4738,-4.0483,1 +-2.9821,4.1986,-0.5898,-3.9642,1 +-1.4628,-1.5706,2.4357,0.49826,1 +-1.7101,-8.7903,7.9735,-0.45475,1 +-1.5572,-9.8808,8.1088,-1.0806,1 +0.74428,-3.7723,1.6131,1.5754,1 +2.0177,1.7982,-2.9581,0.2099,1 +1.164,3.913,-4.5544,-3.8672,1 +-4.3667,6.0692,0.57208,-5.4668,1 +-2.5919,-1.0553,3.8949,0.77757,1 +-1.8046,-6.8141,6.7019,1.1681,1 +-0.71868,-5.7154,3.8298,1.0233,1 +1.4378,0.66837,-2.0267,1.0271,1 +2.1943,4.5503,-4.976,-2.7254,1 +0.7376,4.8525,-4.7986,-5.6659,1 +-5.637,8.1261,0.13081,-5.0142,1 +-3.0193,1.7775,0.73745,-0.45346,1 +-1.6706,-2.09,1.584,0.71162,1 +-0.1269,-1.1505,-0.95138,0.57843,1 +1.2198,2.0982,-3.1954,0.12843,1 +1.4501,3.6067,-4.0557,-1.5966,1 +-0.40857,3.0977,-2.9607,-2.6892,1 +-3.8952,3.8157,-0.31304,-3.8194,1 +-6.3679,8.0102,0.4247,-3.2207,1 +-4.1429,2.7749,0.68261,-0.71984,1 +-2.6864,-0.097265,0.61663,0.061192,1 +-1.0555,0.79459,-1.6968,-0.46768,1 +-0.29858,2.4769,-2.9512,-0.66165,1 +-0.49948,1.7734,-2.2469,-0.68104,1 +-1.9881,0.99945,-0.28562,-0.70044,1 +-1.9389,1.5706,0.045979,-1.122,1 +-1.4375,-1.8624,4.026,0.55127,1 +-3.1875,-7.5756,11.8678,-0.57889,1 +-4.6765,-5.6636,10.969,-0.33449,1 +-2.0285,3.8468,-0.63435,-1.175,1 +0.26637,0.73252,-0.67891,0.03533,1 +-1.7589,-6.4624,8.4773,0.31981,1 +-3.5985,-13.6593,17.6052,-2.4927,1 +-3.3582,-7.2404,11.4419,-0.57113,1 +-2.3629,-0.10554,1.9336,1.1358,1 +-2.1802,3.3791,-1.2256,-2.6621,1 +-0.40951,-0.15521,0.060545,-0.088807,1 +-2.2918,-7.257,7.9597,0.9211,1 +-4.0214,-12.8006,15.6199,-0.95647,1 +-3.3884,-8.215,10.3315,0.98187,1 +-2.0046,-0.49457,1.333,1.6543,1 +-1.7063,2.7956,-2.378,-2.3491,1 +-1.6386,3.3584,-1.7302,-3.5646,1 +-0.41645,0.32487,-0.33617,-0.36036,1 +-1.5877,-6.6072,5.8022,0.31593,1 +-2.5961,-9.349,9.7942,-0.28018,1 +-1.5228,-6.4789,5.7568,0.87325,1 +-0.53072,-0.097265,-0.21793,1.0426,1 +-0.49081,2.8452,-3.6436,-3.1004,1 +-6.5773,6.8017,0.85483,-7.5344,1 +-2.4621,2.7645,-0.62578,-2.8573,1 +-1.3995,-1.9162,2.5154,0.59912,1 +-2.3221,-9.3304,9.233,-0.79871,1 +-3.73,-12.9723,12.9817,-2.684,1 +-1.6988,-7.1163,5.7902,0.16723,1 +-0.26654,-0.64562,-0.42014,0.89136,1 +0.33325,3.3108,-4.5081,-4.012,1 +-4.2091,4.7283,-0.49126,-5.2159,1 +-2.3142,-0.68494,1.9833,-0.44829,1 +-2.4835,-7.4494,6.8964,-0.64484,1 +-2.7611,-10.5099,9.0239,-1.9547,1 +-0.36025,-4.449,2.1067,0.94308,1 +1.0117,0.9022,-2.3506,0.42714,1 +0.96708,3.8426,-4.9314,-4.1323,1 +-5.2049,7.259,0.070827,-7.3004,1 +-3.3203,-0.02691,2.9618,-0.44958,1 +-2.565,-5.7899,6.0122,0.046968,1 +-1.5951,-6.572,4.7689,-0.94354,1 +0.7049,0.17174,-1.7859,0.36119,1 +1.7331,3.9544,-4.7412,-2.5017,1 +0.6818,4.8504,-5.2133,-6.1043,1 +-6.3364,9.2848,0.014275,-6.7844,1 +-3.8053,2.4273,0.6809,-1.0871,1 +-2.1979,-2.1252,1.7151,0.45171,1 +-0.87874,-2.2121,-0.051701,0.099985,1 +0.74067,1.7299,-3.1963,-0.1457,1 +0.98296,3.4226,-3.9692,-1.7116,1 +-0.3489,3.1929,-3.4054,-3.1832,1 +-3.8552,3.5219,-0.38415,-3.8608,1 +-6.9599,8.9931,0.2182,-4.572,1 +-4.7462,3.1205,1.075,-1.2966,1 +-3.2051,-0.14279,0.97565,0.045675,1 +-1.7549,-0.080711,-0.75774,-0.3707,1 +-0.59587,2.4811,-2.8673,-0.89828,1 +-0.89542,2.0279,-2.3652,-1.2746,1 +-2.0754,1.2767,-0.64206,-1.2642,1 +-3.2778,1.8023,0.1805,-2.3931,1 +-2.2183,-1.254,2.9986,0.36378,1 +-3.5895,-6.572,10.5251,-0.16381,1 +-5.0477,-5.8023,11.244,-0.3901,1 +-3.5741,3.944,-0.07912,-2.1203,1 +-0.7351,1.7361,-1.4938,-1.1582,1 +-2.2617,-4.7428,6.3489,0.11162,1 +-4.244,-13.0634,17.1116,-2.8017,1 +-4.0218,-8.304,12.555,-1.5099,1 +-3.0201,-0.67253,2.7056,0.85774,1 +-2.4941,3.5447,-1.3721,-2.8483,1 +-0.83121,0.039307,0.05369,-0.23105,1 +-2.5665,-6.8824,7.5416,0.70774,1 +-4.4018,-12.9371,15.6559,-1.6806,1 +-3.7573,-8.2916,10.3032,0.38059,1 +-2.4725,-0.40145,1.4855,1.1189,1 +-1.9725,2.8825,-2.3086,-2.3724,1 +-2.0149,3.6874,-1.9385,-3.8918,1 +-0.82053,0.65181,-0.48869,-0.52716,1 +-1.7886,-6.3486,5.6154,0.42584,1 +-2.9138,-9.4711,9.7668,-0.60216,1 +-1.8343,-6.5907,5.6429,0.54998,1 +-0.8734,-0.033118,-0.20165,0.55774,1 +-0.70346,2.957,-3.5947,-3.1457,1 +-6.7387,6.9879,0.67833,-7.5887,1 +-2.7723,3.2777,-0.9351,-3.1457,1 +-1.6641,-1.3678,1.997,0.52283,1 +-2.4349,-9.2497,8.9922,-0.50001,1 +-3.793,-12.7095,12.7957,-2.825,1 +-1.9551,-6.9756,5.5383,-0.12889,1 +-0.69078,-0.50077,-0.35417,0.47498,1 +0.025013,3.3998,-4.4327,-4.2655,1 +-4.3967,4.9601,-0.64892,-5.4719,1 +-2.456,-0.24418,1.4041,-0.45863,1 +-2.62,-6.8555,6.2169,-0.62285,1 +-2.9662,-10.3257,8.784,-2.1138,1 +-0.71494,-4.4448,2.2241,0.49826,1 +0.6005,0.99945,-2.2126,0.097399,1 +0.61652,3.8944,-4.7275,-4.3948,1 +-5.4414,7.2363,0.10938,-7.5642,1 +-3.5798,0.45937,2.3457,-0.45734,1 +-2.7769,-5.6967,5.9179,0.37671,1 +-1.8356,-6.7562,5.0585,-0.55044,1 +0.30081,0.17381,-1.7542,0.48921,1 +1.3403,4.1323,-4.7018,-2.5987,1 +0.26877,4.987,-5.1508,-6.3913,1 +-6.5235,9.6014,-0.25392,-6.9642,1 +-4.0679,2.4955,0.79571,-1.1039,1 +-2.564,-1.7051,1.5026,0.32757,1 +-1.3414,-1.9162,-0.15538,-0.11984,1 +0.23874,2.0879,-3.3522,-0.66553,1 +0.6212,3.6771,-4.0771,-2.0711,1 +-0.77848,3.4019,-3.4859,-3.5569,1 +-4.1244,3.7909,-0.6532,-4.1802,1 +-7.0421,9.2,0.25933,-4.6832,1 +-4.9462,3.5716,0.82742,-1.4957,1 +-3.5359,0.30417,0.6569,-0.2957,1 +-2.0662,0.16967,-1.0054,-0.82975,1 +-0.88728,2.808,-3.1432,-1.2035,1 +-1.0941,2.3072,-2.5237,-1.4453,1 +-2.4458,1.6285,-0.88541,-1.4802,1 +-3.551,1.8955,0.1865,-2.4409,1 +-2.2811,-0.85669,2.7185,0.044382,1 +-3.6053,-5.974,10.0916,-0.82846,1 +-5.0676,-5.1877,10.4266,-0.86725,1 +-3.9204,4.0723,-0.23678,-2.1151,1 +-1.1306,1.8458,-1.3575,-1.3806,1 +-2.4561,-4.5566,6.4534,-0.056479,1 +-4.4775,-13.0303,17.0834,-3.0345,1 +-4.1958,-8.1819,12.1291,-1.6017,1 +-3.38,-0.7077,2.5325,0.71808,1 +-2.4365,3.6026,-1.4166,-2.8948,1 +-0.77688,0.13036,-0.031137,-0.35389,1 +-2.7083,-6.8266,7.5339,0.59007,1 +-4.5531,-12.5854,15.4417,-1.4983,1 +-3.8894,-7.8322,9.8208,0.47498,1 +-2.5084,-0.22763,1.488,1.2069,1 +-2.1652,3.0211,-2.4132,-2.4241,1 +-1.8974,3.5074,-1.7842,-3.8491,1 +-0.62043,0.5587,-0.38587,-0.66423,1 +-1.8387,-6.301,5.6506,0.19567,1 +-3,-9.1566,9.5766,-0.73018,1 +-1.9116,-6.1603,5.606,0.48533,1 +-1.005,0.084831,-0.2462,0.45688,1 +-0.87834,3.257,-3.6778,-3.2944,1 +-6.651,6.7934,0.68604,-7.5887,1 +-2.5463,3.1101,-0.83228,-3.0358,1 +-1.4377,-1.432,2.1144,0.42067,1 +-2.4554,-9.0407,8.862,-0.86983,1 +-3.9411,-12.8792,13.0597,-3.3125,1 +-2.1241,-6.8969,5.5992,-0.47156,1 +-0.74324,-0.32902,-0.42785,0.23317,1 +-0.071503,3.7412,-4.5415,-4.2526,1 +-4.2333,4.9166,-0.49212,-5.3207,1 +-2.3675,-0.43663,1.692,-0.43018,1 +-2.5526,-7.3625,6.9255,-0.66811,1 +-3.0986,-10.4602,8.9717,-2.3427,1 +-0.89809,-4.4862,2.2009,0.50731,1 +0.56232,1.0015,-2.2726,-0.0060486,1 +0.53936,3.8944,-4.8166,-4.3418,1 +-5.3012,7.3915,0.029699,-7.3987,1 +-3.3553,0.35591,2.6473,-0.37846,1 +-2.7908,-5.7133,5.953,0.45946,1 +-1.9983,-6.6072,4.8254,-0.41984,1 +0.15423,0.11794,-1.6823,0.59524,1 +1.208,4.0744,-4.7635,-2.6129,1 +0.2952,4.8856,-5.149,-6.2323,1 +-6.4247,9.5311,0.022844,-6.8517,1 +-3.9933,2.6218,0.62863,-1.1595,1 +-2.659,-1.6058,1.3647,0.16464,1 +-1.4094,-2.1252,-0.10397,-0.19225,1 +0.11032,1.9741,-3.3668,-0.65259,1 +0.52374,3.644,-4.0746,-1.9909,1 +-0.76794,3.4598,-3.4405,-3.4276,1 +-3.9698,3.6812,-0.60008,-4.0133,1 +-7.0364,9.2931,0.16594,-4.5396,1 +-4.9447,3.3005,1.063,-1.444,1 +-3.5933,0.22968,0.7126,-0.3332,1 +-2.1674,0.12415,-1.0465,-0.86208,1 +-0.9607,2.6963,-3.1226,-1.3121,1 +-1.0802,2.1996,-2.5862,-1.2759,1 +-2.3277,1.4381,-0.82114,-1.2862,1 +-3.7244,1.9037,-0.035421,-2.5095,1 +-2.5724,-0.95602,2.7073,-0.16639,1 +-3.9297,-6.0816,10.0958,-1.0147,1 +-5.2943,-5.1463,10.3332,-1.1181,1 +-3.8953,4.0392,-0.3019,-2.1836,1 +-1.2244,1.7485,-1.4801,-1.4181,1 +-2.6406,-4.4159,5.983,-0.13924,1 +-4.6338,-12.7509,16.7166,-3.2168,1 +-4.2887,-7.8633,11.8387,-1.8978,1 +-3.3458,-0.50491,2.6328,0.53705,1 +-1.1188,3.3357,-1.3455,-1.9573,1 +0.55939,-0.3104,0.18307,0.44653,1 +-1.5078,-7.3191,7.8981,1.2289,1 +-3.506,-12.5667,15.1606,-0.75216,1 +-2.9498,-8.273,10.2646,1.1629,1 +-1.6029,-0.38903,1.62,1.9103,1 +-1.2667,2.8183,-2.426,-1.8862,1 +-0.49281,3.0605,-1.8356,-2.834,1 +0.66365,-0.045533,-0.18794,0.23447,1 +-0.72068,-6.7583,5.8408,0.62369,1 +-1.9966,-9.5001,9.682,-0.12889,1 +-0.97325,-6.4168,5.6026,1.0323,1 +-0.025314,-0.17383,-0.11339,1.2198,1 +0.062525,2.9301,-3.5467,-2.6737,1 +-5.525,6.3258,0.89768,-6.6241,1 +-1.2943,2.6735,-0.84085,-2.0323,1 +-0.24037,-1.7837,2.135,1.2418,1 +-1.3968,-9.6698,9.4652,-0.34872,1 +-2.9672,-13.2869,13.4727,-2.6271,1 +-1.1005,-7.2508,6.0139,0.36895,1 +0.22432,-0.52147,-0.40386,1.2017,1 +0.90407,3.3708,-4.4987,-3.6965,1 +-2.8619,4.5193,-0.58123,-4.2629,1 +-1.0833,-0.31247,1.2815,0.41291,1 +-1.5681,-7.2446,6.5537,-0.1276,1 +-2.0545,-10.8679,9.4926,-1.4116,1 +0.2346,-4.5152,2.1195,1.4448,1 +1.581,0.86909,-2.3138,0.82412,1 +1.5514,3.8013,-4.9143,-3.7483,1 +-4.1479,7.1225,-0.083404,-6.4172,1 +-2.2625,-0.099335,2.8127,0.48662,1 +-1.7479,-5.823,5.8699,1.212,1 +-0.95923,-6.7128,4.9857,0.32886,1 +1.3451,0.23589,-1.8785,1.3258,1 +2.2279,4.0951,-4.8037,-2.1112,1 +1.2572,4.8731,-5.2861,-5.8741,1 +-5.3857,9.1214,-0.41929,-5.9181,1 +-2.9786,2.3445,0.52667,-0.40173,1 +-1.5851,-2.1562,1.7082,0.9017,1 +-0.21888,-2.2038,-0.0954,0.56421,1 +1.3183,1.9017,-3.3111,0.065071,1 +1.4896,3.4288,-4.0309,-1.4259,1 +0.11592,3.2219,-3.4302,-2.8457,1 +-3.3924,3.3564,-0.72004,-3.5233,1 +-6.1632,8.7096,-0.21621,-3.6345,1 +-4.0786,2.9239,0.87026,-0.65389,1 +-2.5899,-0.3911,0.93452,0.42972,1 +-1.0116,-0.19038,-0.90597,0.003003,1 +0.066129,2.4914,-2.9401,-0.62156,1 +-0.24745,1.9368,-2.4697,-0.80518,1 +-1.5732,1.0636,-0.71232,-0.8388,1 +-2.1668,1.5933,0.045122,-1.678,1 +-1.1667,-1.4237,2.9241,0.66119,1 +-2.8391,-6.63,10.4849,-0.42113,1 +-4.5046,-5.8126,10.8867,-0.52846,1 +-2.41,3.7433,-0.40215,-1.2953,1 +0.40614,1.3492,-1.4501,-0.55949,1 +-1.3887,-4.8773,6.4774,0.34179,1 +-3.7503,-13.4586,17.5932,-2.7771,1 +-3.5637,-8.3827,12.393,-1.2823,1 +-2.5419,-0.65804,2.6842,1.1952,1 \ No newline at end of file diff --git a/doc/pub/week47/ipynb/DataFiles/cancer.dot b/doc/pub/week47/ipynb/DataFiles/cancer.dot new file mode 100644 index 000000000..5b4b48a9b --- /dev/null +++ b/doc/pub/week47/ipynb/DataFiles/cancer.dot @@ -0,0 +1,57 @@ +digraph Tree { +node [shape=box, style="filled, rounded", color="black", fontname=helvetica] ; +edge [fontname=helvetica] ; +0 [label="worst perimeter <= 106.05\ngini = 0.465\nsamples = 426\nvalue = [[269, 157]\n[157, 269]]", fillcolor="#e5813908"] ; +1 [label="worst concave points <= 0.159\ngini = 0.067\nsamples = 259\nvalue = [[250, 9]\n[9, 250]]", fillcolor="#e58139db"] ; +0 -> 1 [labeldistance=2.5, labelangle=45, headlabel="True"] ; +2 [label="worst concave points <= 0.135\ngini = 0.031\nsamples = 253\nvalue = [[249, 4]\n[4, 249]]", fillcolor="#e58139ee"] ; +1 -> 2 ; +3 [label="radius error <= 0.643\ngini = 0.008\nsamples = 242\nvalue = [[241, 1]\n[1, 241]]", fillcolor="#e58139fb"] ; +2 -> 3 ; +4 [label="gini = 0.0\nsamples = 239\nvalue = [[239, 0]\n[0, 239]]", fillcolor="#e58139ff"] ; +3 -> 4 ; +5 [label="worst symmetry <= 0.208\ngini = 0.444\nsamples = 3\nvalue = [[2, 1]\n[1, 2]]", fillcolor="#e5813913"] ; +3 -> 5 ; +6 [label="gini = 0.0\nsamples = 1\nvalue = [[0, 1]\n[1, 0]]", fillcolor="#e58139ff"] ; +5 -> 6 ; +7 [label="gini = 0.0\nsamples = 2\nvalue = [[2, 0]\n[0, 2]]", fillcolor="#e58139ff"] ; +5 -> 7 ; +8 [label="worst texture <= 29.455\ngini = 0.397\nsamples = 11\nvalue = [[8, 3]\n[3, 8]]", fillcolor="#e581392c"] ; +2 -> 8 ; +9 [label="gini = 0.0\nsamples = 8\nvalue = [[8, 0]\n[0, 8]]", fillcolor="#e58139ff"] ; +8 -> 9 ; +10 [label="gini = 0.0\nsamples = 3\nvalue = [[0, 3]\n[3, 0]]", fillcolor="#e58139ff"] ; +8 -> 10 ; +11 [label="mean texture <= 16.22\ngini = 0.278\nsamples = 6\nvalue = [[1, 5]\n[5, 1]]", fillcolor="#e581396b"] ; +1 -> 11 ; +12 [label="gini = 0.0\nsamples = 1\nvalue = [[1, 0]\n[0, 1]]", fillcolor="#e58139ff"] ; +11 -> 12 ; +13 [label="gini = 0.0\nsamples = 5\nvalue = [[0, 5]\n[5, 0]]", fillcolor="#e58139ff"] ; +11 -> 13 ; +14 [label="worst texture <= 20.645\ngini = 0.202\nsamples = 167\nvalue = [[19, 148]\n[148, 19]]", fillcolor="#e5813994"] ; +0 -> 14 [labeldistance=2.5, labelangle=-45, headlabel="False"] ; +15 [label="worst radius <= 17.74\ngini = 0.375\nsamples = 16\nvalue = [[12, 4]\n[4, 12]]", fillcolor="#e5813938"] ; +14 -> 15 ; +16 [label="gini = 0.0\nsamples = 11\nvalue = [[11, 0]\n[0, 11]]", fillcolor="#e58139ff"] ; +15 -> 16 ; +17 [label="mean texture <= 13.745\ngini = 0.32\nsamples = 5\nvalue = [[1, 4]\n[4, 1]]", fillcolor="#e5813955"] ; +15 -> 17 ; +18 [label="gini = 0.0\nsamples = 1\nvalue = [[1, 0]\n[0, 1]]", fillcolor="#e58139ff"] ; +17 -> 18 ; +19 [label="gini = 0.0\nsamples = 4\nvalue = [[0, 4]\n[4, 0]]", fillcolor="#e58139ff"] ; +17 -> 19 ; +20 [label="mean concave points <= 0.049\ngini = 0.088\nsamples = 151\nvalue = [[7, 144]\n[144, 7]]", fillcolor="#e58139d0"] ; +14 -> 20 ; +21 [label="concave points error <= 0.01\ngini = 0.48\nsamples = 15\nvalue = [[6, 9]\n[9, 6]]", fillcolor="#e5813900"] ; +20 -> 21 ; +22 [label="gini = 0.0\nsamples = 9\nvalue = [[0, 9]\n[9, 0]]", fillcolor="#e58139ff"] ; +21 -> 22 ; +23 [label="gini = 0.0\nsamples = 6\nvalue = [[6, 0]\n[0, 6]]", fillcolor="#e58139ff"] ; +21 -> 23 ; +24 [label="worst smoothness <= 0.096\ngini = 0.015\nsamples = 136\nvalue = [[1, 135]\n[135, 1]]", fillcolor="#e58139f7"] ; +20 -> 24 ; +25 [label="gini = 0.0\nsamples = 1\nvalue = [[1, 0]\n[0, 1]]", fillcolor="#e58139ff"] ; +24 -> 25 ; +26 [label="gini = 0.0\nsamples = 135\nvalue = [[0, 135]\n[135, 0]]", fillcolor="#e58139ff"] ; +24 -> 26 ; +} \ No newline at end of file diff --git a/doc/pub/week47/ipynb/DataFiles/cancer.png b/doc/pub/week47/ipynb/DataFiles/cancer.png new file mode 100644 index 000000000..2ceb5e1f8 Binary files /dev/null and b/doc/pub/week47/ipynb/DataFiles/cancer.png differ diff --git a/doc/pub/week47/ipynb/DataFiles/ensembleoverview.png b/doc/pub/week47/ipynb/DataFiles/ensembleoverview.png new file mode 100644 index 000000000..dce581ee6 Binary files /dev/null and b/doc/pub/week47/ipynb/DataFiles/ensembleoverview.png differ diff --git a/doc/pub/week47/ipynb/DataFiles/grades.csv b/doc/pub/week47/ipynb/DataFiles/grades.csv new file mode 100644 index 000000000..eda2ed442 --- /dev/null +++ b/doc/pub/week47/ipynb/DataFiles/grades.csv @@ -0,0 +1,11 @@ +Grade Trend,Hours slept,Hours Studied,Grade +1,0,1,1 +0,1,0,0 +1,0,1,1 +1,1,1,1 +0,0,1,0 +1,0,0,0 +0,1,1,0 +0,0,1,0 +1,0,0,0 +1,1,1,1 diff --git a/doc/pub/week47/ipynb/DataFiles/grades.csv~ b/doc/pub/week47/ipynb/DataFiles/grades.csv~ new file mode 100644 index 000000000..45ca1350d --- /dev/null +++ b/doc/pub/week47/ipynb/DataFiles/grades.csv~ @@ -0,0 +1,11 @@ +Grade Trend,Hours slept,Hours Studied,Grade +1 , 0 , 1 , 1 +0 , 1 , 0 , 0 +1 , 0 , 1 , 1 +1 , 1 , 1 , 1 +0 , 0 , 1 , 0 +1 , 0 , 0 , 0 +0 , 1 , 1 , 0 +0 , 0 , 1 , 0 +1 , 0 , 0 , 0 +1 , 1 , 1 , 1 diff --git a/doc/pub/week47/ipynb/DataFiles/ride.csv b/doc/pub/week47/ipynb/DataFiles/ride.csv new file mode 100644 index 000000000..d03d4ca16 --- /dev/null +++ b/doc/pub/week47/ipynb/DataFiles/ride.csv @@ -0,0 +1,15 @@ +Outlook,Temperature,Humidity,Wind,Ride +0,0,0,0,0 +0,0,0,1,1 +1,0,0,0,1 +2,1,0,0,1 +2,2,1,0,1 +2,2,1,1,0 +1,2,1,1,1 +0,1,0,0,0 +0,2,1,0,1 +2,1,1,0,1 +0,1,1,1,1 +1,1,0,1,1 +1,0,1,0,1 +2,1,0,1,0 diff --git a/doc/pub/week47/ipynb/DataFiles/ride.csv~ b/doc/pub/week47/ipynb/DataFiles/ride.csv~ new file mode 100644 index 000000000..7a908c5f7 --- /dev/null +++ b/doc/pub/week47/ipynb/DataFiles/ride.csv~ @@ -0,0 +1,15 @@ +Outlook,Temperature,Humidity,Wind,Ride +Sunny,Hot,High,Weak,0 +Sunny,Hot,High,Strong,1 +Overcast,Hot,High,Weak,1 +Rain,Mild,High,Weak,1 +Rain,Cool,Normal,Weak,1 +Rain,Cool,Normal,Strong,0 +Overcast,Cool,Normal,Strong,1 +Sunny,Mild,High,Weak,0 +Sunny,Cool,Normal,Weak,1 +Rain,Mild,Normal,Weak,1 +Sunny,Mild,Normal,Strong,1 +Overcast,Mild,High,Strong,1 +Overcast,Hot,Normal,Weak,1 +Rain,Mild,High,Strong,0 diff --git a/doc/pub/week47/ipynb/DataFiles/ride.dot b/doc/pub/week47/ipynb/DataFiles/ride.dot new file mode 100644 index 000000000..50aaa7638 --- /dev/null +++ b/doc/pub/week47/ipynb/DataFiles/ride.dot @@ -0,0 +1,13 @@ +digraph Tree { +node [shape=box, style="filled, rounded", color="black", fontname=helvetica] ; +edge [fontname=helvetica] ; +0 [label="X[7] <= 0.5\ngini = 0.48\nsamples = 15\nvalue = [4, 10, 1]", fillcolor="#39e5818b"] ; +1 [label="X[1] <= 0.5\ngini = 0.408\nsamples = 14\nvalue = [4, 10, 0]", fillcolor="#39e58199"] ; +0 -> 1 [labeldistance=2.5, labelangle=45, headlabel="True"] ; +2 [label="gini = 0.48\nsamples = 10\nvalue = [4, 6, 0]", fillcolor="#39e58155"] ; +1 -> 2 ; +3 [label="gini = 0.0\nsamples = 4\nvalue = [0, 4, 0]", fillcolor="#39e581ff"] ; +1 -> 3 ; +4 [label="gini = 0.0\nsamples = 1\nvalue = [0, 0, 1]", fillcolor="#8139e5ff"] ; +0 -> 4 [labeldistance=2.5, labelangle=-45, headlabel="False"] ; +} \ No newline at end of file diff --git a/doc/pub/week47/ipynb/DataFiles/rideclass.csv b/doc/pub/week47/ipynb/DataFiles/rideclass.csv new file mode 100644 index 000000000..a0831ac7e --- /dev/null +++ b/doc/pub/week47/ipynb/DataFiles/rideclass.csv @@ -0,0 +1,15 @@ +Day,Outlook,Temperature,Humidity,Wind,Ride +1,Sunny,Hot,High,Weak,0 +2,Sunny,Hot,High,Strong,1 +3,Overcast,Hot,High,Weak,1 +4,Rain,Mild,High,Weak,1 +5,Rain,Cool,Normal,Weak,1 +6,Rain,Cool,Normal,Strong,0 +7,Overcast,Cool,Normal,Strong,1 +8,Sunny,Mild,High,Weak,0 +9,Sunny,Cool,Normal,Weak,1 +10,Rain,Mild,Normal,Weak,1 +11,Sunny,Mild,Normal,Strong,1 +12,Overcast,Mild,High,Strong,1 +13,Overcast,Hot,Normal,Weak,1 +14,Rain,Mild,High,Strong,0 diff --git a/doc/pub/week47/ipynb/DataFiles/zoo.csv b/doc/pub/week47/ipynb/DataFiles/zoo.csv new file mode 100644 index 000000000..ca71f7d21 --- /dev/null +++ b/doc/pub/week47/ipynb/DataFiles/zoo.csv @@ -0,0 +1,101 @@ +aardvark,1,0,0,1,0,0,1,1,1,1,0,0,4,0,0,1,1 +antelope,1,0,0,1,0,0,0,1,1,1,0,0,4,1,0,1,1 +bass,0,0,1,0,0,1,1,1,1,0,0,1,0,1,0,0,4 +bear,1,0,0,1,0,0,1,1,1,1,0,0,4,0,0,1,1 +boar,1,0,0,1,0,0,1,1,1,1,0,0,4,1,0,1,1 +buffalo,1,0,0,1,0,0,0,1,1,1,0,0,4,1,0,1,1 +calf,1,0,0,1,0,0,0,1,1,1,0,0,4,1,1,1,1 +carp,0,0,1,0,0,1,0,1,1,0,0,1,0,1,1,0,4 +catfish,0,0,1,0,0,1,1,1,1,0,0,1,0,1,0,0,4 +cavy,1,0,0,1,0,0,0,1,1,1,0,0,4,0,1,0,1 +cheetah,1,0,0,1,0,0,1,1,1,1,0,0,4,1,0,1,1 +chicken,0,1,1,0,1,0,0,0,1,1,0,0,2,1,1,0,2 +chub,0,0,1,0,0,1,1,1,1,0,0,1,0,1,0,0,4 +clam,0,0,1,0,0,0,1,0,0,0,0,0,0,0,0,0,7 +crab,0,0,1,0,0,1,1,0,0,0,0,0,4,0,0,0,7 +crayfish,0,0,1,0,0,1,1,0,0,0,0,0,6,0,0,0,7 +crow,0,1,1,0,1,0,1,0,1,1,0,0,2,1,0,0,2 +deer,1,0,0,1,0,0,0,1,1,1,0,0,4,1,0,1,1 +dogfish,0,0,1,0,0,1,1,1,1,0,0,1,0,1,0,1,4 +dolphin,0,0,0,1,0,1,1,1,1,1,0,1,0,1,0,1,1 +dove,0,1,1,0,1,0,0,0,1,1,0,0,2,1,1,0,2 +duck,0,1,1,0,1,1,0,0,1,1,0,0,2,1,0,0,2 +elephant,1,0,0,1,0,0,0,1,1,1,0,0,4,1,0,1,1 +flamingo,0,1,1,0,1,0,0,0,1,1,0,0,2,1,0,1,2 +flea,0,0,1,0,0,0,0,0,0,1,0,0,6,0,0,0,6 +frog,0,0,1,0,0,1,1,1,1,1,0,0,4,0,0,0,5 +frog,0,0,1,0,0,1,1,1,1,1,1,0,4,0,0,0,5 +fruitbat,1,0,0,1,1,0,0,1,1,1,0,0,2,1,0,0,1 +giraffe,1,0,0,1,0,0,0,1,1,1,0,0,4,1,0,1,1 +girl,1,0,0,1,0,0,1,1,1,1,0,0,2,0,1,1,1 +gnat,0,0,1,0,1,0,0,0,0,1,0,0,6,0,0,0,6 +goat,1,0,0,1,0,0,0,1,1,1,0,0,4,1,1,1,1 +gorilla,1,0,0,1,0,0,0,1,1,1,0,0,2,0,0,1,1 +gull,0,1,1,0,1,1,1,0,1,1,0,0,2,1,0,0,2 +haddock,0,0,1,0,0,1,0,1,1,0,0,1,0,1,0,0,4 +hamster,1,0,0,1,0,0,0,1,1,1,0,0,4,1,1,0,1 +hare,1,0,0,1,0,0,0,1,1,1,0,0,4,1,0,0,1 +hawk,0,1,1,0,1,0,1,0,1,1,0,0,2,1,0,0,2 +herring,0,0,1,0,0,1,1,1,1,0,0,1,0,1,0,0,4 +honeybee,1,0,1,0,1,0,0,0,0,1,1,0,6,0,1,0,6 +housefly,1,0,1,0,1,0,0,0,0,1,0,0,6,0,0,0,6 +kiwi,0,1,1,0,0,0,1,0,1,1,0,0,2,1,0,0,2 +ladybird,0,0,1,0,1,0,1,0,0,1,0,0,6,0,0,0,6 +lark,0,1,1,0,1,0,0,0,1,1,0,0,2,1,0,0,2 +leopard,1,0,0,1,0,0,1,1,1,1,0,0,4,1,0,1,1 +lion,1,0,0,1,0,0,1,1,1,1,0,0,4,1,0,1,1 +lobster,0,0,1,0,0,1,1,0,0,0,0,0,6,0,0,0,7 +lynx,1,0,0,1,0,0,1,1,1,1,0,0,4,1,0,1,1 +mink,1,0,0,1,0,1,1,1,1,1,0,0,4,1,0,1,1 +mole,1,0,0,1,0,0,1,1,1,1,0,0,4,1,0,0,1 +mongoose,1,0,0,1,0,0,1,1,1,1,0,0,4,1,0,1,1 +moth,1,0,1,0,1,0,0,0,0,1,0,0,6,0,0,0,6 +newt,0,0,1,0,0,1,1,1,1,1,0,0,4,1,0,0,5 +octopus,0,0,1,0,0,1,1,0,0,0,0,0,8,0,0,1,7 +opossum,1,0,0,1,0,0,1,1,1,1,0,0,4,1,0,0,1 +oryx,1,0,0,1,0,0,0,1,1,1,0,0,4,1,0,1,1 +ostrich,0,1,1,0,0,0,0,0,1,1,0,0,2,1,0,1,2 +parakeet,0,1,1,0,1,0,0,0,1,1,0,0,2,1,1,0,2 +penguin,0,1,1,0,0,1,1,0,1,1,0,0,2,1,0,1,2 +pheasant,0,1,1,0,1,0,0,0,1,1,0,0,2,1,0,0,2 +pike,0,0,1,0,0,1,1,1,1,0,0,1,0,1,0,1,4 +piranha,0,0,1,0,0,1,1,1,1,0,0,1,0,1,0,0,4 +pitviper,0,0,1,0,0,0,1,1,1,1,1,0,0,1,0,0,3 +platypus,1,0,1,1,0,1,1,0,1,1,0,0,4,1,0,1,1 +polecat,1,0,0,1,0,0,1,1,1,1,0,0,4,1,0,1,1 +pony,1,0,0,1,0,0,0,1,1,1,0,0,4,1,1,1,1 +porpoise,0,0,0,1,0,1,1,1,1,1,0,1,0,1,0,1,1 +puma,1,0,0,1,0,0,1,1,1,1,0,0,4,1,0,1,1 +pussycat,1,0,0,1,0,0,1,1,1,1,0,0,4,1,1,1,1 +raccoon,1,0,0,1,0,0,1,1,1,1,0,0,4,1,0,1,1 +reindeer,1,0,0,1,0,0,0,1,1,1,0,0,4,1,1,1,1 +rhea,0,1,1,0,0,0,1,0,1,1,0,0,2,1,0,1,2 +scorpion,0,0,0,0,0,0,1,0,0,1,1,0,8,1,0,0,7 +seahorse,0,0,1,0,0,1,0,1,1,0,0,1,0,1,0,0,4 +seal,1,0,0,1,0,1,1,1,1,1,0,1,0,0,0,1,1 +sealion,1,0,0,1,0,1,1,1,1,1,0,1,2,1,0,1,1 +seasnake,0,0,0,0,0,1,1,1,1,0,1,0,0,1,0,0,3 +seawasp,0,0,1,0,0,1,1,0,0,0,1,0,0,0,0,0,7 +skimmer,0,1,1,0,1,1,1,0,1,1,0,0,2,1,0,0,2 +skua,0,1,1,0,1,1,1,0,1,1,0,0,2,1,0,0,2 +slowworm,0,0,1,0,0,0,1,1,1,1,0,0,0,1,0,0,3 +slug,0,0,1,0,0,0,0,0,0,1,0,0,0,0,0,0,7 +sole,0,0,1,0,0,1,0,1,1,0,0,1,0,1,0,0,4 +sparrow,0,1,1,0,1,0,0,0,1,1,0,0,2,1,0,0,2 +squirrel,1,0,0,1,0,0,0,1,1,1,0,0,2,1,0,0,1 +starfish,0,0,1,0,0,1,1,0,0,0,0,0,5,0,0,0,7 +stingray,0,0,1,0,0,1,1,1,1,0,1,1,0,1,0,1,4 +swan,0,1,1,0,1,1,0,0,1,1,0,0,2,1,0,1,2 +termite,0,0,1,0,0,0,0,0,0,1,0,0,6,0,0,0,6 +toad,0,0,1,0,0,1,0,1,1,1,0,0,4,0,0,0,5 +tortoise,0,0,1,0,0,0,0,0,1,1,0,0,4,1,0,1,3 +tuatara,0,0,1,0,0,0,1,1,1,1,0,0,4,1,0,0,3 +tuna,0,0,1,0,0,1,1,1,1,0,0,1,0,1,0,1,4 +vampire,1,0,0,1,1,0,0,1,1,1,0,0,2,1,0,0,1 +vole,1,0,0,1,0,0,0,1,1,1,0,0,4,1,0,0,1 +vulture,0,1,1,0,1,0,1,0,1,1,0,0,2,1,0,1,2 +wallaby,1,0,0,1,0,0,0,1,1,1,0,0,2,1,0,1,1 +wasp,1,0,1,0,1,0,0,0,0,1,1,0,6,0,0,0,6 +wolf,1,0,0,1,0,0,1,1,1,1,0,0,4,1,0,1,1 +worm,0,0,1,0,0,0,0,0,0,1,0,0,0,0,0,0,7 +wren,0,1,1,0,1,0,0,0,1,1,0,0,2,1,0,0,2 diff --git a/doc/pub/week47/ipynb/Results/FigureFiles/baggingboot.png b/doc/pub/week47/ipynb/Results/FigureFiles/baggingboot.png new file mode 100644 index 000000000..6b7edb4be Binary files /dev/null and b/doc/pub/week47/ipynb/Results/FigureFiles/baggingboot.png differ diff --git a/doc/pub/week47/ipynb/Results/FigureFiles/gdclassiffiercgain.png b/doc/pub/week47/ipynb/Results/FigureFiles/gdclassiffiercgain.png new file mode 100644 index 000000000..48e9113d9 Binary files /dev/null and b/doc/pub/week47/ipynb/Results/FigureFiles/gdclassiffiercgain.png differ diff --git a/doc/pub/week47/ipynb/Results/FigureFiles/gdclassiffierconfusion.png b/doc/pub/week47/ipynb/Results/FigureFiles/gdclassiffierconfusion.png new file mode 100644 index 000000000..6eeb87be0 Binary files /dev/null and b/doc/pub/week47/ipynb/Results/FigureFiles/gdclassiffierconfusion.png differ diff --git a/doc/pub/week47/ipynb/Results/FigureFiles/gdclassiffierroc.png b/doc/pub/week47/ipynb/Results/FigureFiles/gdclassiffierroc.png new file mode 100644 index 000000000..142ba57ab Binary files /dev/null and b/doc/pub/week47/ipynb/Results/FigureFiles/gdclassiffierroc.png differ diff --git a/doc/pub/week47/ipynb/Results/FigureFiles/gdregression.png b/doc/pub/week47/ipynb/Results/FigureFiles/gdregression.png new file mode 100644 index 000000000..75d3e868e Binary files /dev/null and b/doc/pub/week47/ipynb/Results/FigureFiles/gdregression.png differ diff --git a/doc/pub/week47/ipynb/Results/FigureFiles/xdclassiffierconfusion.png b/doc/pub/week47/ipynb/Results/FigureFiles/xdclassiffierconfusion.png new file mode 100644 index 000000000..b203d1aaf Binary files /dev/null and b/doc/pub/week47/ipynb/Results/FigureFiles/xdclassiffierconfusion.png differ diff --git a/doc/pub/week47/ipynb/Results/FigureFiles/xdclassiffierroc.png b/doc/pub/week47/ipynb/Results/FigureFiles/xdclassiffierroc.png new file mode 100644 index 000000000..f2d294a80 Binary files /dev/null and b/doc/pub/week47/ipynb/Results/FigureFiles/xdclassiffierroc.png differ diff --git a/doc/pub/week47/ipynb/Results/FigureFiles/xgparams.png b/doc/pub/week47/ipynb/Results/FigureFiles/xgparams.png new file mode 100644 index 000000000..3afce4739 Binary files /dev/null and b/doc/pub/week47/ipynb/Results/FigureFiles/xgparams.png differ diff --git a/doc/pub/week47/ipynb/Results/FigureFiles/xgtree.png b/doc/pub/week47/ipynb/Results/FigureFiles/xgtree.png new file mode 100644 index 000000000..54c780906 Binary files /dev/null and b/doc/pub/week47/ipynb/Results/FigureFiles/xgtree.png differ diff --git a/doc/pub/week47/ipynb/week46.ipynb b/doc/pub/week47/ipynb/week46.ipynb new file mode 100644 index 000000000..a89a55843 --- /dev/null +++ b/doc/pub/week47/ipynb/week46.ipynb @@ -0,0 +1,2796 @@ +{ + "cells": [ + { + "cell_type": "markdown", + "id": "0457d6a8", + "metadata": {}, + "source": [ + "\n", + "" + ] + }, + { + "cell_type": "markdown", + "id": "17b8d5b2", + "metadata": {}, + "source": [ + "# Week 46: Decision Trees, Ensemble methods and Random Forests\n", + "**Morten Hjorth-Jensen**, Department of Physics, University of Oslo and Department of Physics and Astronomy and National Superconducting Cyclotron Laboratory, Michigan State University\n", + "\n", + "Date: **Week 46, November 11-15**" + ] + }, + { + "cell_type": "markdown", + "id": "78852c55", + "metadata": {}, + "source": [ + "## Plan for week 46\n", + "\n", + "**Lab sessions on Tuesday and Wednesday.**\n", + "\n", + "1. Work on and discussions of project 3\n", + "\n", + "**Material for the lecture on Monday November 11, 2024.**\n", + "\n", + "Basics of decision trees, classification and regression algorithms and ensemble models \n", + "1. Readings and Videos:\n", + "\n", + "a. Lecture notes atFigure 1:
\n", + "\n", + "\n", + "This tree was produced using the Wisconsin cancer data (discussed here as well, see code examples below) using **Scikit-Learn**'s decision tree classifier. Here we have used the so-called **gini** index (see below) to split the various branches." + ] + }, + { + "cell_type": "markdown", + "id": "f57ac382", + "metadata": {}, + "source": [ + "## General Features\n", + "\n", + "The overarching approach to decision trees is a top-down approach.\n", + "\n", + "* A leaf provides the classification of a given instance.\n", + "\n", + "* A node specifies a test of some attribute of the instance.\n", + "\n", + "* A branch corresponds to a possible values of an attribute.\n", + "\n", + "* An instance is classified by starting at the root node of the tree, testing the attribute specified by this node, then moving down the tree branch corresponding to the value of the attribute in the given example.\n", + "\n", + "This process is then repeated for the subtree rooted at the new\n", + "node." + ] + }, + { + "cell_type": "markdown", + "id": "a9619651", + "metadata": {}, + "source": [ + "## How do we set it up?\n", + "\n", + "In simplified terms, the process of training a decision tree and\n", + "predicting the target features of query instances is as follows:\n", + "\n", + "1. Present a dataset containing of a number of training instances characterized by a number of descriptive features and a target feature\n", + "\n", + "2. Train the decision tree model by continuously splitting the target feature along the values of the descriptive features using a measure of information gain during the training process\n", + "\n", + "3. Grow the tree until we accomplish a stopping criteria create leaf nodes which represent the *predictions* we want to make for new query instances\n", + "\n", + "4. Show query instances to the tree and run down the tree until we arrive at leaf nodes\n", + "\n", + "Then we are essentially done!" + ] + }, + { + "cell_type": "markdown", + "id": "72fccb7b", + "metadata": {}, + "source": [ + "## Decision trees and Regression" + ] + }, + { + "cell_type": "code", + "execution_count": 1, + "id": "53bc6c2a", + "metadata": {}, + "outputs": [], + "source": [ + "%matplotlib inline\n", + "\n", + "import numpy as np\n", + "import matplotlib.pyplot as plt\n", + "from sklearn.preprocessing import PolynomialFeatures\n", + "from sklearn.linear_model import LinearRegression\n", + "\n", + "steps=250\n", + "\n", + "distance=0\n", + "x=0\n", + "distance_list=[]\n", + "steps_list=[]\n", + "while x| Day | Outlook | Temperature | Humidity | Wind | Ride |
|---|---|---|---|---|---|
| 1 | Sunny | Hot | High | Weak | 0 |
| 2 | Sunny | Hot | High | Strong | 1 |
| 3 | Overcast | Hot | High | Weak | 1 |
| 4 | Rain | Mild | High | Weak | 1 |
| 5 | Rain | Cool | Normal | Weak | 1 |
| 6 | Rain | Cool | Normal | Strong | 0 |
| 7 | Overcast | Cool | Normal | Strong | 1 |
| 8 | Sunny | Mild | High | Weak | 0 |
| 9 | Sunny | Cool | Normal | Weak | 1 |
| 10 | Rain | Mild | Normal | Weak | 1 |
| 11 | Sunny | Mild | Normal | Strong | 1 |
| 12 | Overcast | Mild | High | Strong | 1 |
| 13 | Overcast | Hot | Normal | Weak | 1 |
| 14 | Rain | Mild | High | Strong | 0 |
Figure 1:
\n", + "" + ] + }, + { + "cell_type": "markdown", + "id": "a9001c69", + "metadata": {}, + "source": [ + "## Why Voting?\n", + "\n", + "The idea behind boosting, and voting as well can be phrased as follows:\n", + "**Can a group of people somehow arrive at highly\n", + "reasoned decisions, despite the weak judgement of the individual\n", + "members?**\n", + "\n", + "The aim is to create a good classifier by combining several weak classifiers.\n", + "**A weak classifier is a classifier which is able to produce results that are only slightly better than guessing at random.**\n", + "\n", + "The basic approach is to apply repeatedly (in boosting this is done in an iterative way) a weak classifier to modifications of the data.\n", + "In voting we simply apply the law of large numbers while in boosting we give more weight to misclassified data in\n", + "each iteration. \n", + "\n", + "Decision trees play an important role as our weak classifier. They serve as the basic method." + ] + }, + { + "cell_type": "markdown", + "id": "95bf38e7", + "metadata": {}, + "source": [ + "## Tossing coins\n", + "\n", + "The simplest case is a so-called voting ensemble. To illustrate this,\n", + "think of yourself tossing coins with a biased outcome of 51 per cent\n", + "for heads and 49% for tails. With only few tosses,\n", + "you may not clearly see this distribution for heads and tails. However, after some\n", + "thousands of tosses, there will be a clear majority of heads. With 2000 tosses\n", + "you should see approximately 1020 heads and 980 tails.\n", + "\n", + "We can then state that the outcome is a clear majority of heads. If\n", + "you do this ten thousand times, it is easy to see that there is a 97%\n", + "likelihood of a majority of heads.\n", + "\n", + "Another example would be to collect all polls before an\n", + "election. Different polls may show different likelihoods for a\n", + "candidate winning with say a majority of the popular vote. The majority vote\n", + "would then consist in many polls indicating that this candidate will\n", + "actually win.\n", + "\n", + "The example here shows how we can implement the coin tossing case,\n", + "clealry demostrating that after some tosses we see the [law of large](https://en.wikipedia.org/wiki/Law_of_large_numbers)\n", + "numbers kicking in." + ] + }, + { + "cell_type": "markdown", + "id": "aba2349a", + "metadata": {}, + "source": [ + "## Standard imports first" + ] + }, + { + "cell_type": "code", + "execution_count": 13, + "id": "62312165", + "metadata": {}, + "outputs": [], + "source": [ + "# Common imports\n", + "from IPython.display import Image \n", + "from pydot import graph_from_dot_data\n", + "import pandas as pd\n", + "import numpy as np\n", + "import matplotlib.pyplot as plt\n", + "from sklearn.tree import DecisionTreeClassifier\n", + "from sklearn.model_selection import train_test_split\n", + "from sklearn.tree import export_graphviz\n", + "from sklearn.preprocessing import StandardScaler, OneHotEncoder\n", + "from sklearn.compose import ColumnTransformer\n", + "from IPython.display import Image \n", + "from pydot import graph_from_dot_data\n", + "import os\n", + "\n", + "# Where to save the figures and data files\n", + "PROJECT_ROOT_DIR = \"Results\"\n", + "FIGURE_ID = \"Results/FigureFiles\"\n", + "DATA_ID = \"DataFiles/\"\n", + "\n", + "if not os.path.exists(PROJECT_ROOT_DIR):\n", + " os.mkdir(PROJECT_ROOT_DIR)\n", + "\n", + "if not os.path.exists(FIGURE_ID):\n", + " os.makedirs(FIGURE_ID)\n", + "\n", + "if not os.path.exists(DATA_ID):\n", + " os.makedirs(DATA_ID)\n", + "\n", + "def image_path(fig_id):\n", + " return os.path.join(FIGURE_ID, fig_id)\n", + "\n", + "def data_path(dat_id):\n", + " return os.path.join(DATA_ID, dat_id)\n", + "\n", + "def save_fig(fig_id):\n", + " plt.savefig(image_path(fig_id) + \".png\", format='png')" + ] + }, + { + "cell_type": "markdown", + "id": "7307b530", + "metadata": {}, + "source": [ + "## Simple Voting Example, head or tail" + ] + }, + { + "cell_type": "code", + "execution_count": 14, + "id": "b9bac328", + "metadata": {}, + "outputs": [], + "source": [ + "\n", + "# Common imports\n", + "import numpy as np\n", + "import matplotlib\n", + "import matplotlib.pyplot as plt\n", + "from matplotlib.colors import ListedColormap\n", + "plt.rcParams['axes.labelsize'] = 14\n", + "plt.rcParams['xtick.labelsize'] = 12\n", + "plt.rcParams['ytick.labelsize'] = 12\n", + "\n", + "heads_proba = 0.51\n", + "coin_tosses = (np.random.rand(10000, 10) < heads_proba).astype(np.int32)\n", + "cumulative_heads_ratio = np.cumsum(coin_tosses, axis=0) / np.arange(1, 10001).reshape(-1, 1)\n", + "plt.figure(figsize=(8,3.5))\n", + "plt.plot(cumulative_heads_ratio)\n", + "plt.plot([0, 10000], [0.51, 0.51], \"k--\", linewidth=2, label=\"51%\")\n", + "plt.plot([0, 10000], [0.5, 0.5], \"k-\", label=\"50%\")\n", + "plt.xlabel(\"Number of coin tosses\")\n", + "plt.ylabel(\"Heads ratio\")\n", + "plt.legend(loc=\"lower right\")\n", + "plt.axis([0, 10000, 0.42, 0.58])\n", + "save_fig(\"votingsimple\")\n", + "plt.show()" + ] + }, + { + "cell_type": "markdown", + "id": "8ab7d717", + "metadata": {}, + "source": [ + "## Using the Voting Classifier\n", + "\n", + "We can use the voting classifier on other data sets, here the exciting binary case of two distinct objects using the make moons functionality of **Scikit-Learn**." + ] + }, + { + "cell_type": "code", + "execution_count": 15, + "id": "a4b9c350", + "metadata": {}, + "outputs": [], + "source": [ + "from sklearn.model_selection import train_test_split\n", + "from sklearn.datasets import make_moons\n", + "\n", + "X, y = make_moons(n_samples=500, noise=0.30, random_state=42)\n", + "X_train, X_test, y_train, y_test = train_test_split(X, y, random_state=42)\n", + "\n", + "from sklearn.ensemble import RandomForestClassifier\n", + "from sklearn.ensemble import VotingClassifier\n", + "from sklearn.linear_model import LogisticRegression\n", + "from sklearn.svm import SVC\n", + "\n", + "log_clf = LogisticRegression(solver=\"liblinear\", random_state=42)\n", + "rnd_clf = RandomForestClassifier(n_estimators=10, random_state=42)\n", + "svm_clf = SVC(gamma=\"auto\", random_state=42)\n", + "\n", + "voting_clf = VotingClassifier(\n", + " estimators=[('lr', log_clf), ('rf', rnd_clf), ('svc', svm_clf)],\n", + " voting='hard')\n", + "\n", + "voting_clf.fit(X_train, y_train)\n", + "\n", + "from sklearn.metrics import accuracy_score\n", + "\n", + "for clf in (log_clf, rnd_clf, svm_clf, voting_clf):\n", + " clf.fit(X_train, y_train)\n", + " y_pred = clf.predict(X_test)\n", + " print(clf.__class__.__name__, accuracy_score(y_test, y_pred))\n", + "\n", + "log_clf = LogisticRegression(solver=\"liblinear\", random_state=42)\n", + "rnd_clf = RandomForestClassifier(n_estimators=10, random_state=42)\n", + "svm_clf = SVC(gamma=\"auto\", probability=True, random_state=42)\n", + "voting_clf = VotingClassifier(\n", + " estimators=[('lr', log_clf), ('rf', rnd_clf), ('svc', svm_clf)],\n", + " voting='soft')\n", + "voting_clf.fit(X_train, y_train)\n", + "\n", + "from sklearn.metrics import accuracy_score\n", + "\n", + "for clf in (log_clf, rnd_clf, svm_clf, voting_clf):\n", + " clf.fit(X_train, y_train)\n", + " y_pred = clf.predict(X_test)\n", + " print(clf.__class__.__name__, accuracy_score(y_test, y_pred))" + ] + }, + { + "cell_type": "markdown", + "id": "8b42c140", + "metadata": {}, + "source": [ + "## Voting and Bagging" + ] + }, + { + "cell_type": "code", + "execution_count": 16, + "id": "2d91b7c4", + "metadata": {}, + "outputs": [], + "source": [ + "from sklearn.model_selection import train_test_split\n", + "from sklearn.datasets import make_moons\n", + "\n", + "X, y = make_moons(n_samples=500, noise=0.30, random_state=42)\n", + "X_train, X_test, y_train, y_test = train_test_split(X, y, random_state=42)\n", + "from sklearn.ensemble import RandomForestClassifier\n", + "from sklearn.ensemble import VotingClassifier\n", + "from sklearn.linear_model import LogisticRegression\n", + "from sklearn.svm import SVC\n", + "\n", + "log_clf = LogisticRegression(random_state=42)\n", + "rnd_clf = RandomForestClassifier(random_state=42)\n", + "svm_clf = SVC(random_state=42)\n", + "\n", + "voting_clf = VotingClassifier(\n", + " estimators=[('lr', log_clf), ('rf', rnd_clf), ('svc', svm_clf)],\n", + " voting='hard')\n", + "voting_clf.fit(X_train, y_train)" + ] + }, + { + "cell_type": "code", + "execution_count": 17, + "id": "841ca35e", + "metadata": {}, + "outputs": [], + "source": [ + "from sklearn.metrics import accuracy_score\n", + "\n", + "for clf in (log_clf, rnd_clf, svm_clf, voting_clf):\n", + " clf.fit(X_train, y_train)\n", + " y_pred = clf.predict(X_test)\n", + " print(clf.__class__.__name__, accuracy_score(y_test, y_pred))" + ] + }, + { + "cell_type": "code", + "execution_count": 18, + "id": "f6fb547e", + "metadata": {}, + "outputs": [], + "source": [ + "log_clf = LogisticRegression(random_state=42)\n", + "rnd_clf = RandomForestClassifier(random_state=42)\n", + "svm_clf = SVC(probability=True, random_state=42)\n", + "\n", + "voting_clf = VotingClassifier(\n", + " estimators=[('lr', log_clf), ('rf', rnd_clf), ('svc', svm_clf)],\n", + " voting='soft')\n", + "voting_clf.fit(X_train, y_train)" + ] + }, + { + "cell_type": "code", + "execution_count": 19, + "id": "21d21aef", + "metadata": {}, + "outputs": [], + "source": [ + "from sklearn.metrics import accuracy_score\n", + "\n", + "for clf in (log_clf, rnd_clf, svm_clf, voting_clf):\n", + " clf.fit(X_train, y_train)\n", + " y_pred = clf.predict(X_test)\n", + " print(clf.__class__.__name__, accuracy_score(y_test, y_pred))" + ] + }, + { + "cell_type": "markdown", + "id": "8a08fc9a", + "metadata": {}, + "source": [ + "## Bagging\n", + "\n", + "The **plain** decision trees suffer from high\n", + "variance. This means that if we split the training data into two parts\n", + "at random, and fit a decision tree to both halves, the results that we\n", + "get could be quite different. In contrast, a procedure with low\n", + "variance will yield similar results if applied repeatedly to distinct\n", + "data sets; linear regression tends to have low variance, if the ratio\n", + "of $n$ to $p$ is moderately large. \n", + "\n", + "**Bootstrap aggregation**, or just **bagging**, is a\n", + "general-purpose procedure for reducing the variance of a statistical\n", + "learning method." + ] + }, + { + "cell_type": "markdown", + "id": "189d48af", + "metadata": {}, + "source": [ + "## More bagging\n", + "\n", + "Bagging typically results in improved accuracy\n", + "over prediction using a single tree. Unfortunately, however, it can be\n", + "difficult to interpret the resulting model. Recall that one of the\n", + "advantages of decision trees is the attractive and easily interpreted\n", + "diagram that results.\n", + "\n", + "However, when we bag a large number of trees, it is no longer\n", + "possible to represent the resulting statistical learning procedure\n", + "using a single tree, and it is no longer clear which variables are\n", + "most important to the procedure. Thus, bagging improves prediction\n", + "accuracy at the expense of interpretability. Although the collection\n", + "of bagged trees is much more difficult to interpret than a single\n", + "tree, one can obtain an overall summary of the importance of each\n", + "predictor using the MSE (for bagging regression trees) or the Gini\n", + "index (for bagging classification trees). In the case of bagging\n", + "regression trees, we can record the total amount that the MSE is\n", + "decreased due to splits over a given predictor, averaged over all $B$ possible\n", + "trees. A large value indicates an important predictor. Similarly, in\n", + "the context of bagging classification trees, we can add up the total\n", + "amount that the Gini index is decreased by splits over a given\n", + "predictor, averaged over all $B$ trees." + ] + }, + { + "cell_type": "markdown", + "id": "b6cfe02e", + "metadata": {}, + "source": [ + "## Making your own Bootstrap: Changing the Level of the Decision Tree\n", + "\n", + "Let us bring up our good old boostrap example from the linear regression lectures. We change the linerar regression algorithm with\n", + "a decision tree wth different depths and perform a bootstrap aggregate (in this case we perform as many bootstraps as data points $n$)." + ] + }, + { + "cell_type": "code", + "execution_count": 20, + "id": "609bd17f", + "metadata": {}, + "outputs": [], + "source": [ + "\n", + "import matplotlib.pyplot as plt\n", + "import numpy as np\n", + "from sklearn.model_selection import train_test_split\n", + "from sklearn.pipeline import make_pipeline\n", + "from sklearn.utils import resample\n", + "from sklearn.tree import DecisionTreeRegressor\n", + "\n", + "n = 100\n", + "n_boostraps = 100\n", + "maxdepth = 8\n", + "\n", + "# Make data set.\n", + "x = np.linspace(-3, 3, n).reshape(-1, 1)\n", + "y = np.exp(-x**2) + 1.5 * np.exp(-(x-2)**2)+ np.random.normal(0, 0.1, x.shape)\n", + "error = np.zeros(maxdepth)\n", + "bias = np.zeros(maxdepth)\n", + "variance = np.zeros(maxdepth)\n", + "polydegree = np.zeros(maxdepth)\n", + "X_train, X_test, y_train, y_test = train_test_split(x, y, test_size=0.2)\n", + "\n", + "from sklearn.preprocessing import StandardScaler\n", + "scaler = StandardScaler()\n", + "scaler.fit(X_train)\n", + "X_train_scaled = scaler.transform(X_train)\n", + "X_test_scaled = scaler.transform(X_test)\n", + "\n", + "# we produce a simple tree first as benchmark\n", + "simpletree = DecisionTreeRegressor(max_depth=3) \n", + "simpletree.fit(X_train_scaled, y_train)\n", + "simpleprediction = simpletree.predict(X_test_scaled)\n", + "for degree in range(1,maxdepth):\n", + " model = DecisionTreeRegressor(max_depth=degree) \n", + " y_pred = np.empty((y_test.shape[0], n_boostraps))\n", + " for i in range(n_boostraps):\n", + " x_, y_ = resample(X_train_scaled, y_train)\n", + " model.fit(x_, y_)\n", + " y_pred[:, i] = model.predict(X_test_scaled)#.ravel()\n", + "\n", + " polydegree[degree] = degree\n", + " error[degree] = np.mean( np.mean((y_test - y_pred)**2, axis=1, keepdims=True) )\n", + " bias[degree] = np.mean( (y_test - np.mean(y_pred, axis=1, keepdims=True))**2 )\n", + " variance[degree] = np.mean( np.var(y_pred, axis=1, keepdims=True) )\n", + " print('Polynomial degree:', degree)\n", + " print('Error:', error[degree])\n", + " print('Bias^2:', bias[degree])\n", + " print('Var:', variance[degree])\n", + " print('{} >= {} + {} = {}'.format(error[degree], bias[degree], variance[degree], bias[degree]+variance[degree]))\n", + " \n", + "mse_simpletree= np.mean( np.mean((y_test - simpleprediction)**2))\n", + "print(\"Simple tree:\",mse_simpletree)\n", + "plt.xlim(1,maxdepth)\n", + "plt.plot(polydegree, error, label='MSE')\n", + "plt.plot(polydegree, bias, label='bias')\n", + "plt.plot(polydegree, variance, label='Variance')\n", + "plt.legend()\n", + "save_fig(\"baggingboot\")\n", + "plt.show()" + ] + }, + { + "cell_type": "markdown", + "id": "bd392364", + "metadata": {}, + "source": [ + "## Random forests\n", + "\n", + "Random forests provide an improvement over bagged trees by way of a\n", + "small tweak that decorrelates the trees. \n", + "\n", + "As in bagging, we build a\n", + "number of decision trees on bootstrapped training samples. But when\n", + "building these decision trees, each time a split in a tree is\n", + "considered, a random sample of $m$ predictors is chosen as split\n", + "candidates from the full set of $p$ predictors. The split is allowed to\n", + "use only one of those $m$ predictors. \n", + "\n", + "A fresh sample of $m$ predictors is\n", + "taken at each split, and typically we choose" + ] + }, + { + "cell_type": "markdown", + "id": "5498b4fd", + "metadata": {}, + "source": [ + "$$\n", + "m\\approx \\sqrt{p}.\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "356ce239", + "metadata": {}, + "source": [ + "In building a random forest, at\n", + "each split in the tree, the algorithm is not even allowed to consider\n", + "a majority of the available predictors. \n", + "\n", + "The reason for this is rather clever. Suppose that there is one very\n", + "strong predictor in the data set, along with a number of other\n", + "moderately strong predictors. Then in the collection of bagged\n", + "variable importance random forest trees, most or all of the trees will\n", + "use this strong predictor in the top split. Consequently, all of the\n", + "bagged trees will look quite similar to each other. Hence the\n", + "predictions from the bagged trees will be highly correlated.\n", + "Unfortunately, averaging many highly correlated quantities does not\n", + "lead to as large of a reduction in variance as averaging many\n", + "uncorrelated quantities. In particular, this means that bagging will\n", + "not lead to a substantial reduction in variance over a single tree in\n", + "this setting." + ] + }, + { + "cell_type": "markdown", + "id": "de8ed0de", + "metadata": {}, + "source": [ + "## Random Forest Algorithm\n", + "The algorithm described here can be applied to both classification and regression problems.\n", + "\n", + "We will grow of forest of say $B$ trees.\n", + "1. For $b=1:B$\n", + "\n", + " * Draw a bootstrap sample from the training data organized in our $\\boldsymbol{X}$ matrix.\n", + "\n", + " * We grow then a random forest tree $T_b$ based on the bootstrapped data by repeating the steps outlined till we reach the maximum node size is reached\n", + "\n", + "1. we select $m \\le p$ variables at random from the $p$ predictors/features\n", + "\n", + "2. pick the best split point among the $m$ features using for example the CART algorithm and create a new node\n", + "\n", + "3. split the node into daughter nodes\n", + "\n", + "4. Output then the ensemble of trees $\\{T_b\\}_1^{B}$ and make predictions for either a regression type of problem or a classification type of problem." + ] + }, + { + "cell_type": "markdown", + "id": "b54f39b2", + "metadata": {}, + "source": [ + "## Random Forests Compared with other Methods on the Cancer Data" + ] + }, + { + "cell_type": "code", + "execution_count": 21, + "id": "ebf1724c", + "metadata": {}, + "outputs": [], + "source": [ + "import matplotlib.pyplot as plt\n", + "import numpy as np\n", + "from sklearn.model_selection import train_test_split \n", + "from sklearn.datasets import load_breast_cancer\n", + "from sklearn.svm import SVC\n", + "from sklearn.linear_model import LogisticRegression\n", + "from sklearn.tree import DecisionTreeClassifier\n", + "from sklearn.ensemble import BaggingClassifier\n", + "\n", + "# Load the data\n", + "cancer = load_breast_cancer()\n", + "\n", + "X_train, X_test, y_train, y_test = train_test_split(cancer.data,cancer.target,random_state=0)\n", + "print(X_train.shape)\n", + "print(X_test.shape)\n", + "#define methods\n", + "# Logistic Regression\n", + "logreg = LogisticRegression(solver='lbfgs')\n", + "# Support vector machine\n", + "svm = SVC(gamma='auto', C=100)\n", + "# Decision Trees\n", + "deep_tree_clf = DecisionTreeClassifier(max_depth=None)\n", + "#Scale the data\n", + "from sklearn.preprocessing import StandardScaler\n", + "scaler = StandardScaler()\n", + "scaler.fit(X_train)\n", + "X_train_scaled = scaler.transform(X_train)\n", + "X_test_scaled = scaler.transform(X_test)\n", + "# Logistic Regression\n", + "logreg.fit(X_train_scaled, y_train)\n", + "print(\"Test set accuracy Logistic Regression with scaled data: {:.2f}\".format(logreg.score(X_test_scaled,y_test)))\n", + "# Support Vector Machine\n", + "svm.fit(X_train_scaled, y_train)\n", + "print(\"Test set accuracy SVM with scaled data: {:.2f}\".format(logreg.score(X_test_scaled,y_test)))\n", + "# Decision Trees\n", + "deep_tree_clf.fit(X_train_scaled, y_train)\n", + "print(\"Test set accuracy with Decision Trees and scaled data: {:.2f}\".format(deep_tree_clf.score(X_test_scaled,y_test)))\n", + "\n", + "\n", + "from sklearn.ensemble import RandomForestClassifier\n", + "from sklearn.preprocessing import LabelEncoder\n", + "from sklearn.model_selection import cross_validate\n", + "# Data set not specificied\n", + "#Instantiate the model with 500 trees and entropy as splitting criteria\n", + "Random_Forest_model = RandomForestClassifier(n_estimators=500,criterion=\"entropy\")\n", + "Random_Forest_model.fit(X_train_scaled, y_train)\n", + "#Cross validation\n", + "accuracy = cross_validate(Random_Forest_model,X_test_scaled,y_test,cv=10)['test_score']\n", + "print(accuracy)\n", + "print(\"Test set accuracy with Random Forests and scaled data: {:.2f}\".format(Random_Forest_model.score(X_test_scaled,y_test)))\n", + "\n", + "\n", + "import scikitplot as skplt\n", + "y_pred = Random_Forest_model.predict(X_test_scaled)\n", + "skplt.metrics.plot_confusion_matrix(y_test, y_pred, normalize=True)\n", + "plt.show()\n", + "y_probas = Random_Forest_model.predict_proba(X_test_scaled)\n", + "skplt.metrics.plot_roc(y_test, y_probas)\n", + "plt.show()\n", + "skplt.metrics.plot_cumulative_gain(y_test, y_probas)\n", + "plt.show()" + ] + }, + { + "cell_type": "markdown", + "id": "3c629901", + "metadata": {}, + "source": [ + "Recall that the cumulative gains curve shows the percentage of the\n", + "overall number of cases in a given category *gained* by targeting a\n", + "percentage of the total number of cases.\n", + "\n", + "Similarly, the receiver operating characteristic curve, or ROC curve,\n", + "displays the diagnostic ability of a binary classifier system as its\n", + "discrimination threshold is varied. It plots the true positive rate against the false positive rate." + ] + }, + { + "cell_type": "markdown", + "id": "6717213a", + "metadata": {}, + "source": [ + "## Compare Bagging on Trees with Random Forests" + ] + }, + { + "cell_type": "code", + "execution_count": 22, + "id": "2ee05d8e", + "metadata": {}, + "outputs": [], + "source": [ + "bag_clf = BaggingClassifier(\n", + " DecisionTreeClassifier(splitter=\"random\", max_leaf_nodes=16, random_state=42),\n", + " n_estimators=500, max_samples=1.0, bootstrap=True, n_jobs=-1, random_state=42)" + ] + }, + { + "cell_type": "code", + "execution_count": 23, + "id": "b8c99064", + "metadata": {}, + "outputs": [], + "source": [ + "bag_clf.fit(X_train, y_train)\n", + "y_pred = bag_clf.predict(X_test)\n", + "from sklearn.ensemble import RandomForestClassifier\n", + "rnd_clf = RandomForestClassifier(n_estimators=500, max_leaf_nodes=16, n_jobs=-1, random_state=42)\n", + "rnd_clf.fit(X_train, y_train)\n", + "y_pred_rf = rnd_clf.predict(X_test)\n", + "np.sum(y_pred == y_pred_rf) / len(y_pred)" + ] + }, + { + "cell_type": "markdown", + "id": "99d07558", + "metadata": {}, + "source": [ + "## Boosting, a Bird's Eye View\n", + "\n", + "The basic idea is to combine weak classifiers in order to create a good\n", + "classifier. With a weak classifier we often intend a classifier which\n", + "produces results which are only slightly better than we would get by\n", + "random guesses.\n", + "\n", + "This is done by applying in an iterative way a weak (or a standard\n", + "classifier like decision trees) to modify the data. In each iteration\n", + "we emphasize those observations which are misclassified by weighting\n", + "them with a factor." + ] + }, + { + "cell_type": "markdown", + "id": "4ea2cf75", + "metadata": {}, + "source": [ + "## What is boosting? Additive Modelling/Iterative Fitting\n", + "\n", + "Boosting is a way of fitting an additive expansion in a set of\n", + "elementary basis functions like for example some simple polynomials.\n", + "Assume for example that we have a function" + ] + }, + { + "cell_type": "markdown", + "id": "eacb1356", + "metadata": {}, + "source": [ + "$$\n", + "f_M(x) = \\sum_{i=1}^M \\beta_m b(x;\\gamma_m),\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "e014f7e3", + "metadata": {}, + "source": [ + "where $\\beta_m$ are the expansion parameters to be determined in a\n", + "minimization process and $b(x;\\gamma_m)$ are some simple functions of\n", + "the multivariable parameter $x$ which is characterized by the\n", + "parameters $\\gamma_m$.\n", + "\n", + "As an example, consider the Sigmoid function we used in logistic\n", + "regression. In that case, we can translate the function\n", + "$b(x;\\gamma_m)$ into the Sigmoid function" + ] + }, + { + "cell_type": "markdown", + "id": "d3673f5c", + "metadata": {}, + "source": [ + "$$\n", + "\\sigma(t) = \\frac{1}{1+\\exp{(-t)}},\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "121752e8", + "metadata": {}, + "source": [ + "where $t=\\gamma_0+\\gamma_1 x$ and the parameters $\\gamma_0$ and\n", + "$\\gamma_1$ were determined by the Logistic Regression fitting\n", + "algorithm.\n", + "\n", + "As another example, consider the cost function we defined for linear regression" + ] + }, + { + "cell_type": "markdown", + "id": "7d3c5c69", + "metadata": {}, + "source": [ + "$$\n", + "C(\\boldsymbol{y},\\boldsymbol{f}) = \\frac{1}{n} \\sum_{i=0}^{n-1}(y_i-f(x_i))^2.\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "bd1bc84e", + "metadata": {}, + "source": [ + "In this case the function $f(x)$ was replaced by the design matrix\n", + "$\\boldsymbol{X}$ and the unknown linear regression parameters $\\boldsymbol{\\beta}$,\n", + "that is $\\boldsymbol{f}=\\boldsymbol{X}\\boldsymbol{\\beta}$. In linear regression we can \n", + "simply invert a matrix and obtain the parameters $\\beta$ by" + ] + }, + { + "cell_type": "markdown", + "id": "7135b2c5", + "metadata": {}, + "source": [ + "$$\n", + "\\boldsymbol{\\beta}=\\left(\\boldsymbol{X}^T\\boldsymbol{X}\\right)^{-1}\\boldsymbol{X}^T\\boldsymbol{y}.\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "a3531473", + "metadata": {}, + "source": [ + "In iterative fitting or additive modeling, we minimize the cost function with respect to the parameters $\\beta_m$ and $\\gamma_m$." + ] + }, + { + "cell_type": "markdown", + "id": "4e5e0786", + "metadata": {}, + "source": [ + "## Iterative Fitting, Regression and Squared-error Cost Function\n", + "\n", + "The way we proceed is as follows (here we specialize to the squared-error cost function)\n", + "\n", + "1. Establish a cost function, here $C(\\boldsymbol{y},\\boldsymbol{f}) = \\frac{1}{n} \\sum_{i=0}^{n-1}(y_i-f_M(x_i))^2$ with $f_M(x) = \\sum_{i=1}^M \\beta_m b(x;\\gamma_m)$.\n", + "\n", + "2. Initialize with a guess $f_0(x)$. It could be one or even zero or some random numbers.\n", + "\n", + "3. For $m=1:M$\n", + "\n", + "a. minimize $\\sum_{i=0}^{n-1}(y_i-f_{m-1}(x_i)-\\beta b(x;\\gamma))^2$ wrt $\\gamma$ and $\\beta$\n", + "\n", + "b. This gives the optimal values $\\beta_m$ and $\\gamma_m$\n", + "\n", + "c. Determine then the new values $f_m(x)=f_{m-1}(x) +\\beta_m b(x;\\gamma_m)$\n", + "\n", + "We could use any of the algorithms we have discussed till now. If we\n", + "use trees, $\\gamma$ parameterizes the split variables and split points\n", + "at the internal nodes, and the predictions at the terminal nodes." + ] + }, + { + "cell_type": "markdown", + "id": "bd71f33b", + "metadata": {}, + "source": [ + "## Squared-Error Example and Iterative Fitting\n", + "\n", + "To better understand what happens, let us develop the steps for the iterative fitting using the above squared error function.\n", + "\n", + "For simplicity we assume also that our functions $b(x;\\gamma)=1+\\gamma x$. \n", + "\n", + "This means that for every iteration $m$, we need to optimize" + ] + }, + { + "cell_type": "markdown", + "id": "029244f7", + "metadata": {}, + "source": [ + "$$\n", + "(\\beta_m,\\gamma_m) = \\mathrm{argmin}_{\\beta,\\lambda}\\hspace{0.1cm} \\sum_{i=0}^{n-1}(y_i-f_{m-1}(x_i)-\\beta b(x;\\gamma))^2=\\sum_{i=0}^{n-1}(y_i-f_{m-1}(x_i)-\\beta(1+\\gamma x_i))^2.\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "f72c0d58", + "metadata": {}, + "source": [ + "We start our iteration by simply setting $f_0(x)=0$. \n", + "Taking the derivatives with respect to $\\beta$ and $\\gamma$ we obtain" + ] + }, + { + "cell_type": "markdown", + "id": "1b7d8540", + "metadata": {}, + "source": [ + "$$\n", + "\\frac{\\partial {\\cal C}}{\\partial \\beta} = -2\\sum_{i}(1+\\gamma x_i)(y_i-\\beta(1+\\gamma x_i))=0,\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "2d35f2df", + "metadata": {}, + "source": [ + "and" + ] + }, + { + "cell_type": "markdown", + "id": "b5859a45", + "metadata": {}, + "source": [ + "$$\n", + "\\frac{\\partial {\\cal C}}{\\partial \\gamma} =-2\\sum_{i}\\beta x_i(y_i-\\beta(1+\\gamma x_i))=0.\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "6329b7c2", + "metadata": {}, + "source": [ + "We can then rewrite these equations as (defining $\\boldsymbol{w}=\\boldsymbol{e}+\\gamma \\boldsymbol{x})$ with $\\boldsymbol{e}$ being the unit vector)" + ] + }, + { + "cell_type": "markdown", + "id": "b7cc50c4", + "metadata": {}, + "source": [ + "$$\n", + "\\gamma \\boldsymbol{w}^T(\\boldsymbol{y}-\\beta\\gamma \\boldsymbol{w})=0,\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "2282bbac", + "metadata": {}, + "source": [ + "which gives us $\\beta = \\boldsymbol{w}^T\\boldsymbol{y}/(\\boldsymbol{w}^T\\boldsymbol{w})$. Similarly we have" + ] + }, + { + "cell_type": "markdown", + "id": "917718aa", + "metadata": {}, + "source": [ + "$$\n", + "\\beta\\gamma \\boldsymbol{x}^T(\\boldsymbol{y}-\\beta(1+\\gamma \\boldsymbol{x}))=0,\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "c66d9b04", + "metadata": {}, + "source": [ + "which leads to $\\gamma =(\\boldsymbol{x}^T\\boldsymbol{y}-\\beta\\boldsymbol{x}^T\\boldsymbol{e})/(\\beta\\boldsymbol{x}^T\\boldsymbol{x})$. Inserting\n", + "for $\\beta$ gives us an equation for $\\gamma$. This is a non-linear equation in the unknown $\\gamma$ and has to be solved numerically. \n", + "\n", + "The solution to these two equations gives us in turn $\\beta_1$ and $\\gamma_1$ leading to the new expression for $f_1(x)$ as\n", + "$f_1(x) = \\beta_1(1+\\gamma_1x)$. Doing this $M$ times results in our final estimate for the function $f$." + ] + }, + { + "cell_type": "markdown", + "id": "341567c5", + "metadata": {}, + "source": [ + "## Iterative Fitting, Classification and AdaBoost\n", + "\n", + "Let us consider a binary classification problem with two outcomes $y_i \\in \\{-1,1\\}$ and $i=0,1,2,\\dots,n-1$ as our set of\n", + "observations. We define a classification function $G(x)$ which produces a prediction taking one or the other of the two values \n", + "$\\{-1,1\\}$.\n", + "\n", + "The error rate of the training sample is then" + ] + }, + { + "cell_type": "markdown", + "id": "52afa6c2", + "metadata": {}, + "source": [ + "$$\n", + "\\mathrm{\\overline{err}}=\\frac{1}{n} \\sum_{i=0}^{n-1} I(y_i\\ne G(x_i)).\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "e02b6199", + "metadata": {}, + "source": [ + "The iterative procedure starts with defining a weak classifier whose\n", + "error rate is barely better than random guessing. The iterative\n", + "procedure in boosting is to sequentially apply a weak\n", + "classification algorithm to repeatedly modified versions of the data\n", + "producing a sequence of weak classifiers $G_m(x)$.\n", + "\n", + "Here we will express our function $f(x)$ in terms of $G(x)$. That is" + ] + }, + { + "cell_type": "markdown", + "id": "03761f49", + "metadata": {}, + "source": [ + "$$\n", + "f_M(x) = \\sum_{i=1}^M \\beta_m b(x;\\gamma_m),\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "63757911", + "metadata": {}, + "source": [ + "will be a function of" + ] + }, + { + "cell_type": "markdown", + "id": "4b0c293a", + "metadata": {}, + "source": [ + "$$\n", + "G_M(x) = \\mathrm{sign} \\sum_{i=1}^M \\alpha_m G_m(x).\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "70122b7c", + "metadata": {}, + "source": [ + "## Adaptive Boosting, AdaBoost\n", + "\n", + "In our iterative procedure we define thus" + ] + }, + { + "cell_type": "markdown", + "id": "a9bd3da3", + "metadata": {}, + "source": [ + "$$\n", + "f_m(x) = f_{m-1}(x)+\\beta_mG_m(x).\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "4a67628f", + "metadata": {}, + "source": [ + "The simplest possible cost function which leads (also simple from a computational point of view) to the AdaBoost algorithm is the\n", + "exponential cost/loss function defined as" + ] + }, + { + "cell_type": "markdown", + "id": "ef2cf02f", + "metadata": {}, + "source": [ + "$$\n", + "C(\\boldsymbol{y},\\boldsymbol{f}) = \\sum_{i=0}^{n-1}\\exp{(-y_i(f_{m-1}(x_i)+\\beta G(x_i))}.\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "b5cf283d", + "metadata": {}, + "source": [ + "We optimize $\\beta$ and $G$ for each value of $m=1:M$ as we did in the regression case.\n", + "This is normally done in two steps. Let us however first rewrite the cost function as" + ] + }, + { + "cell_type": "markdown", + "id": "8c326212", + "metadata": {}, + "source": [ + "$$\n", + "C(\\boldsymbol{y},\\boldsymbol{f}) = \\sum_{i=0}^{n-1}w_i^{m}\\exp{(-y_i\\beta G(x_i))},\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "e2e7dbcb", + "metadata": {}, + "source": [ + "where we have defined $w_i^m= \\exp{(-y_if_{m-1}(x_i))}$." + ] + }, + { + "cell_type": "markdown", + "id": "8418bbcb", + "metadata": {}, + "source": [ + "## Building up AdaBoost\n", + "\n", + "First, for any $\\beta > 0$, we optimize $G$ by setting" + ] + }, + { + "cell_type": "markdown", + "id": "f28e7f81", + "metadata": {}, + "source": [ + "$$\n", + "G_m(x) = \\mathrm{sign} \\sum_{i=0}^{n-1} w_i^m I(y_i \\ne G_(x_i)),\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "87059974", + "metadata": {}, + "source": [ + "which is the classifier that minimizes the weighted error rate in predicting $y$.\n", + "\n", + "We can do this by rewriting" + ] + }, + { + "cell_type": "markdown", + "id": "65dc0f70", + "metadata": {}, + "source": [ + "$$\n", + "\\exp{-(\\beta)}\\sum_{y_i=G(x_i)}w_i^m+\\exp{(\\beta)}\\sum_{y_i\\ne G(x_i)}w_i^m,\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "7a409f3d", + "metadata": {}, + "source": [ + "which can be rewritten as" + ] + }, + { + "cell_type": "markdown", + "id": "7b339df7", + "metadata": {}, + "source": [ + "$$\n", + "(\\exp{(\\beta)}-\\exp{-(\\beta)})\\sum_{i=0}^{n-1}w_i^mI(y_i\\ne G(x_i))+\\exp{(-\\beta)}\\sum_{i=0}^{n-1}w_i^m=0,\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "857ad2e1", + "metadata": {}, + "source": [ + "which leads to" + ] + }, + { + "cell_type": "markdown", + "id": "280d3ff0", + "metadata": {}, + "source": [ + "$$\n", + "\\beta_m = \\frac{1}{2}\\log{\\frac{1-\\mathrm{\\overline{err}}}{\\mathrm{\\overline{err}}}},\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "7f08ccc4", + "metadata": {}, + "source": [ + "where we have redefined the error as" + ] + }, + { + "cell_type": "markdown", + "id": "5f3ef2a9", + "metadata": {}, + "source": [ + "$$\n", + "\\mathrm{\\overline{err}}_m=\\frac{1}{n}\\frac{\\sum_{i=0}^{n-1}w_i^mI(y_i\\ne G(x_i)}{\\sum_{i=0}^{n-1}w_i^m},\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "bb658722", + "metadata": {}, + "source": [ + "which leads to an update of" + ] + }, + { + "cell_type": "markdown", + "id": "8fd114af", + "metadata": {}, + "source": [ + "$$\n", + "f_m(x) = f_{m-1}(x) +\\beta_m G_m(x).\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "79579995", + "metadata": {}, + "source": [ + "This leads to the new weights" + ] + }, + { + "cell_type": "markdown", + "id": "b47cdb72", + "metadata": {}, + "source": [ + "$$\n", + "w_i^{m+1} = w_i^m \\exp{(-y_i\\beta_m G_m(x_i))}\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "f8385ba9", + "metadata": {}, + "source": [ + "## Adaptive boosting: AdaBoost, Basic Algorithm\n", + "\n", + "The algorithm here is rather straightforward. Assume that our weak\n", + "classifier is a decision tree and we consider a binary set of outputs\n", + "with $y_i \\in \\{-1,1\\}$ and $i=0,1,2,\\dots,n-1$ as our set of\n", + "observations. Our design matrix is given in terms of the\n", + "feature/predictor vectors\n", + "$\\boldsymbol{X}=[\\boldsymbol{x}_0\\boldsymbol{x}_1\\dots\\boldsymbol{x}_{p-1}]$. Finally, we define also a\n", + "classifier determined by our data via a function $G(x)$. This function tells us how well we are able to classify our outputs/targets $\\boldsymbol{y}$. \n", + "\n", + "We have already defined the misclassification error $\\mathrm{err}$ as" + ] + }, + { + "cell_type": "markdown", + "id": "3b7343ad", + "metadata": {}, + "source": [ + "$$\n", + "\\mathrm{err}=\\frac{1}{n}\\sum_{i=0}^{n-1}I(y_i\\ne G(x_i)),\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "debdf61a", + "metadata": {}, + "source": [ + "where the function $I()$ is one if we misclassify and zero if we classify correctly." + ] + }, + { + "cell_type": "markdown", + "id": "a836f153", + "metadata": {}, + "source": [ + "## Basic Steps of AdaBoost\n", + "\n", + "With the above definitions we are now ready to set up the algorithm for AdaBoost.\n", + "The basic idea is to set up weights which will be used to scale the correctly classified and the misclassified cases.\n", + "1. We start by initializing all weights to $w_i = 1/n$, with $i=0,1,2,\\dots n-1$. It is easy to see that we must have $\\sum_{i=0}^{n-1}w_i = 1$.\n", + "\n", + "2. We rewrite the misclassification error as" + ] + }, + { + "cell_type": "markdown", + "id": "4b5ec620", + "metadata": {}, + "source": [ + "$$\n", + "\\mathrm{\\overline{err}}_m=\\frac{\\sum_{i=0}^{n-1}w_i^m I(y_i\\ne G(x_i))}{\\sum_{i=0}^{n-1}w_i},\n", + "$$" + ] + }, + { + "cell_type": "markdown", + "id": "18f2330a", + "metadata": {}, + "source": [ + "1. Then we start looping over all attempts at classifying, namely we start an iterative process for $m=1:M$, where $M$ is the final number of classifications. Our given classifier could for example be a plain decision tree.\n", + "\n", + "a. Fit then a given classifier to the training set using the weights $w_i$.\n", + "\n", + "b. Compute then $\\mathrm{err}$ and figure out which events are classified properly and which are classified wrongly.\n", + "\n", + "c. Define a quantity $\\alpha_{m} = \\log{(1-\\mathrm{\\overline{err}}_m)/\\mathrm{\\overline{err}}_m}$\n", + "\n", + "d. Set the new weights to $w_i = w_i\\times \\exp{(\\alpha_m I(y_i\\ne G(x_i)}$.\n", + "\n", + "5. Compute the new classifier $G(x)= \\sum_{i=0}^{n-1}\\alpha_m I(y_i\\ne G(x_i)$.\n", + "\n", + "For the iterations with $m \\le 2$ the weights are modified\n", + "individually at each steps. The observations which were misclassified\n", + "at iteration $m-1$ have a weight which is larger than those which were\n", + "classified properly. As this proceeds, the observations which were\n", + "difficult to classifiy correctly are given a larger influence. Each\n", + "new classification step $m$ is then forced to concentrate on those\n", + "observations that are missed in the previous iterations." + ] + }, + { + "cell_type": "markdown", + "id": "34ed9bbb", + "metadata": {}, + "source": [ + "## AdaBoost Examples\n", + "\n", + "Using **Scikit-Learn** it is easy to apply the adaptive boosting algorithm, as done here." + ] + }, + { + "cell_type": "code", + "execution_count": 24, + "id": "923ecc83", + "metadata": {}, + "outputs": [], + "source": [ + "from sklearn.ensemble import AdaBoostClassifier\n", + "\n", + "ada_clf = AdaBoostClassifier(\n", + " DecisionTreeClassifier(max_depth=1), n_estimators=200,\n", + " algorithm=\"SAMME.R\", learning_rate=0.5, random_state=42)\n", + "ada_clf.fit(X_train, y_train)\n", + "\n", + "from sklearn.ensemble import AdaBoostClassifier\n", + "\n", + "ada_clf = AdaBoostClassifier(\n", + " DecisionTreeClassifier(max_depth=1), n_estimators=200,\n", + " algorithm=\"SAMME.R\", learning_rate=0.5, random_state=42)\n", + "ada_clf.fit(X_train_scaled, y_train)\n", + "y_pred = ada_clf.predict(X_test_scaled)\n", + "skplt.metrics.plot_confusion_matrix(y_test, y_pred, normalize=True)\n", + "plt.show()\n", + "y_probas = ada_clf.predict_proba(X_test_scaled)\n", + "skplt.metrics.plot_roc(y_test, y_probas)\n", + "plt.show()\n", + "skplt.metrics.plot_cumulative_gain(y_test, y_probas)\n", + "plt.show()" + ] + } + ], + "metadata": { + "kernelspec": { + "display_name": "Python 3 (ipykernel)", + "language": "python", + "name": "python3" + }, + "language_info": { + "codemirror_mode": { + "name": "ipython", + "version": 3 + }, + "file_extension": ".py", + "mimetype": "text/x-python", + "name": "python", + "nbconvert_exporter": "python", + "pygments_lexer": "ipython3", + "version": "3.9.15" + } + }, + "nbformat": 4, + "nbformat_minor": 5 +}