@@ -161,7 +196,7 @@ MathJax.Hub.Config({
-
@@ -169,689 +204,101 @@ MathJax.Hub.Config({ -
-Cubic spline interpolation is among one of the most used -methods for interpolating between data points where the arguments -are organized as ascending series. In the library program we supply -such a function, based on the so-called cubic spline method to be -described below. +
-A spline function consists of polynomial pieces defined on -subintervals. The different subintervals are connected via -various continuity relations. - -
-Assume we have at our disposal \( n+1 \) points \( x_0, x_1, \dots x_n \) -arranged so that \( x_0 < x_1 < x_2 < \dots x_{n-1} < x_n \) (such points are called -knots). A spline function \( s \) of degree \( k \) with \( n+1 \) knots is defined -as follows - -
-
-As an example, consider a spline function of degree \( k=1 \) defined as follows +
+The method of steepest descent The basic idea of gradient descent is +that a function \( F(\mathbf{x}) \), +\( \mathbf{x} \equiv (x_1,\cdots,x_n) \), decreases fastest if one goes from \( \bf {x} \) in the +direction of the negative gradient \( -\nabla F(\mathbf{x}) \). + +
+It can be shown that if $$ - s(x)=\begin{bmatrix} s_0(x)=a_0x+b_0 & x\in [x_0, x_1) \\ - s_1(x)=a_1x+b_1 & x\in [x_1, x_2) \\ - \dots & \dots \\ - s_{n-1}(x)=a_{n-1}x+b_{n-1} & x\in - [x_{n-1}, x_n] \end{bmatrix}. +\mathbf{x}_{k+1} = \mathbf{x}_k - \gamma_k \nabla F(\mathbf{x}_k), \ \ \gamma_k > 0 $$ -In this case the polynomial consists of series of straight lines -connected to each other at every endpoint. The number of continuous -derivatives is then \( k-1=0 \), as expected when we deal with straight lines. -Such a polynomial is quite easy to construct given -\( n+1 \) points \( x_0, x_1, \dots x_n \) and their corresponding -function values. -
+for \( \gamma_k \) small enough, then \( F(\mathbf{x}_{k+1}) \leq +F(\mathbf{x}_k) \). This means that for a sufficiently small \( \gamma_k \) +we are always moving towards smaller function values, i.e a minimum. +
+ + +
+The previous observation is the basis of the method of steepest +descent, which is also referred to as just gradient descent (GD). One +starts with an initial guess \( \mathbf{x}_0 \) for a minimum of \( F \) and +compute new approximations according to + +$$ +\mathbf{x}_{k+1} = \mathbf{x}_k - \gamma_k \nabla F(\mathbf{x}_k), \ \ k \geq 0. +$$ + +
+The parameter \( \gamma_k \) is often referred to as the step length or +the learning rate in the context of Machine Learning. + +
+ + +
+Ideally the sequence \( \{ \mathbf{x}_k \}_{k=0} \) converges to a global minimum of the function \( F \). In general we do not know if we are in a global or local minimum. In the special case when \( F \) is a convex function, all local minima are also global minima, so in this case gradient descent can converge to the global solution. The advantage of this scheme is that it is conceptually simple and straightforward to implement. However the method in this form has some severe limitations: + +
+In machine learing we are often faced with non-convex high dimensional cost functions with many local minimum. Since GD is deterministic we will get stuck in a local minimum, if the method converges, unless we have a very good intial guess. This also implies that the scheme is sensitive to the chosen initial condition. + +
+Note that the gradient is a function of \( \mathbf{x} = +(x_1,\cdots,x_n) \) which makes it expensive to compute numerically. + +
+ + +
+GD is sensitive to the choice of learning rate \( \gamma_k \). This is due +to the fact that we are only guaranteed that \( F(\mathbf{x}_{k+1}) \leq +F(\mathbf{x}_k) \) for sufficiently small \( \gamma_k \). The problem is to +determine an optimal learning rate. If the learning rate is chosen to +small the method will take a long to converge and if it is to large we +can experience erratic behavior. + +
+Many of these shortcomings can be alleviated by introducing +randomness. One such method is that of Stochastic Gradient Descent +(SGD), see below
-
-The most commonly used spline function is the one with \( k=3 \), the so-called -cubic spline function. -Assume that we have in adddition to the \( n+1 \) knots a series of -functions values \( y_0=f(x_0), y_1=f(x_1), \dots y_n=f(x_n) \). -By definition, the polynomials \( s_{i-1} \) and \( s_i \) -are thence supposed to interpolate the same point \( i \), that is -$$ - s_{i-1}(x_i)= y_i = s_i(x_i), -$$ - -with \( 1 \le i \le n-1 \). In total we have \( n \) polynomials of the -type -$$ - s_i(x)=a_{i0}+a_{i1}x+a_{i2}x^2+a_{i2}x^3, -$$ - -yielding \( 4n \) coefficients to determine. -
- - -
-Every subinterval provides in addition the \( 2n \) conditions -$$ - y_i = s(x_i), -$$ - -and -$$ - s(x_{i+1})= y_{i+1}, -$$ - -to be fulfilled. If we also assume that \( s' \) and \( s'' \) are continuous, -then -$$ - s'_{i-1}(x_i)= s'_i(x_i), -$$ - -yields \( n-1 \) conditions. Similarly, -$$ - s''_{i-1}(x_i)= s''_i(x_i), -$$ - -results in additional \( n-1 \) conditions. In total we have \( 4n \) coefficients -and \( 4n-2 \) equations to determine them, leaving us with \( 2 \) degrees of -freedom to be determined. -
- - -
-Using the last equation we define two values for the second derivative, namely -$$ - s''_{i}(x_i)= f_i, -$$ - -and -$$ - s''_{i}(x_{i+1})= f_{i+1}, -$$ - -and setting up a straight line between \( f_i \) and \( f_{i+1} \) we have -$$ - s_i''(x) = \frac{f_i}{x_{i+1}-x_i}(x_{i+1}-x)+ - \frac{f_{i+1}}{x_{i+1}-x_i}(x-x_i), -$$ - -and integrating twice one obtains -$$ - s_i(x) = \frac{f_i}{6(x_{i+1}-x_i)}(x_{i+1}-x)^3+ - \frac{f_{i+1}}{6(x_{i+1}-x_i)}(x-x_i)^3 - +c(x-x_i)+d(x_{i+1}-x). -$$ -
- - -
-Using the conditions \( s_i(x_i)=y_i \) and \( s_i(x_{i+1})=y_{i+1} \) -we can in turn determine the constants \( c \) and \( d \) resulting in -$$ -\begin{align} - s_i(x) =&\frac{f_i}{6(x_{i+1}-x_i)}(x_{i+1}-x)^3+ - \frac{f_{i+1}}{6(x_{i+1}-x_i)}(x-x_i)^3 \nonumber \\ - +&(\frac{y_{i+1}}{x_{i+1}-x_i}-\frac{f_{i+1}(x_{i+1}-x_i)}{6}) - (x-x_i)+ - (\frac{y_{i}}{x_{i+1}-x_i}-\frac{f_{i}(x_{i+1}-x_i)}{6}) - (x_{i+1}-x). -\label{_auto1} -\end{align} -$$ -
- - -
-How to determine the values of the second -derivatives \( f_{i} \) and \( f_{i+1} \)? We use the continuity assumption -of the first derivatives -$$ - s'_{i-1}(x_i)= s'_i(x_i), -$$ - -and set \( x=x_i \). Defining \( h_i=x_{i+1}-x_i \) we obtain finally -the following expression -$$ - h_{i-1}f_{i-1}+2(h_{i}+h_{i-1})f_i+h_if_{i+1}= - \frac{6}{h_i}(y_{i+1}-y_i)-\frac{6}{h_{i-1}}(y_{i}-y_{i-1}), -$$ - -and introducing the shorthands \( u_i=2(h_{i}+h_{i-1}) \), -\( v_i=\frac{6}{h_i}(y_{i+1}-y_i)-\frac{6}{h_{i-1}}(y_{i}-y_{i-1}) \), -we can reformulate the problem as a set of linear equations to be -solved through e.g., Gaussian elemination -
- - -
-Gaussian elimination -$$ - \begin{bmatrix} u_1 & h_1 &0 &\dots & & & & \\ - h_1 & u_2 & h_2 &0 &\dots & & & \\ - 0 & h_2 & u_3 & h_3 &0 &\dots & & \\ - \dots& & \dots &\dots &\dots &\dots &\dots & \\ - &\dots & & &0 &h_{n-3} &u_{n-2} &h_{n-2} \\ - & && & &0 &h_{n-2} &u_{n-1} \end{bmatrix} - \begin{bmatrix} f_1 \\ - f_2 \\ - f_3\\ - \dots \\ - f_{n-2} \\ - f_{n-1} \end{bmatrix} = - \begin{bmatrix} v_1 \\ - v_2 \\ - v_3\\ - \dots \\ - v_{n-2}\\ - v_{n-1} \end{bmatrix}. -$$ - -Note that this is a set of tridiagonal equations and can be solved -through only \( O(n) \) operations. -
- - -
-The functions supplied in the program library are spline and splint. -In order to use cubic spline interpolation you need first to call - -
- - -
spline(double x[], double y[], int n, double yp1, double yp2, double y2[])
--This function takes as -input \( x[0,..,n - 1] \) and \( y[0,..,n - 1] \) containing a tabulation -\( y_i = f(x_i) \) with \( x_0 < x_1 < .. < x_{n - 1} \) -together with the -first derivatives of \( f(x) \) at \( x_0 \) and \( x_{n-1} \), respectively. Then the -function returns \( y2[0,..,n-1] \) which contains the second derivatives of -\( f(x_i) \) at each point \( x_i \). \( n \) is the number of points. -This function provides the cubic spline interpolation for all subintervals -and is called only once. -
- - -
-Thereafter, if you wish to make various interpolations, you need to call the function -
- - -
splint(double x[], double y[], double y2a[], int n, double x, double *y)
--which takes as input -the tabulated values \( x[0,..,n - 1] \) and \( y[0,..,n - 1] \) and the output -y2a[0,..,n - 1] from spline. It returns the value \( y \) corresponding -to the point \( x \). -
- - -
-The success of the CG method for finding solutions of non-linear problems is based -on the theory of conjugate gradients for linear systems of equations. It belongs -to the class of iterative methods for solving problems from linear algebra of the type -$$ -\begin{equation*} - \hat{A}\hat{x} = \hat{b}. -\end{equation*} -$$ - -In the iterative process we end up with a problem like - -$$ -\begin{equation*} - \hat{r}= \hat{b}-\hat{A}\hat{x}, -\end{equation*} -$$ - -where \( \hat{r} \) is the so-called residual or error in the iterative process. - -
-When we have found the exact solution, \( \hat{r}=0 \). -
- - -
- -
-The residual is zero when we reach the minimum of the quadratic equation -$$ -\begin{equation*} - P(\hat{x})=\frac{1}{2}\hat{x}^T\hat{A}\hat{x} - \hat{x}^T\hat{b}, -\end{equation*} -$$ - -with the constraint that the matrix \( \hat{A} \) is positive definite and symmetric. -If we search for a minimum of the quantum mechanical variance, then the matrix -\( \hat{A} \), which is called the Hessian, is given by the second-derivative of the function we want to minimize. This quantity is always positive definite. In our case this corresponds normally to the second derivative of the energy. -
- - -
-We seek the minimum of the energy or the variance as function of various variational parameters. -In our case we have thus a function \( f \) whose minimum we are seeking. -In Newton's method we set \( \nabla f = 0 \) and we can thus compute the next iteration point -$$ -\begin{equation*} -\hat{x}-\hat{x}_i=\hat{A}^{-1}\nabla f(\hat{x}_i). -\end{equation*} -$$ - -Subtracting this equation from that of \( \hat{x}_{i+1} \) we have -$$ -\begin{equation*} -\hat{x}_{i+1}-\hat{x}_i=\hat{A}^{-1}(\nabla f(\hat{x}_{i+1})-\nabla f(\hat{x}_i)). -\end{equation*} -$$ -
- - -
-The function \( f \) can be either the energy or the variance. If we choose the energy then we have -$$ -\begin{equation*} -\hat{\alpha}_{i+1}-\hat{\alpha}_i=\hat{A}^{-1}(\nabla E(\hat{\alpha}_{i+1})-\nabla E(\hat{\alpha}_i)). -\end{equation*} -$$ - -In the simple harmonic oscillator model, the gradient and the Hessian \( \hat{A} \) are -$$ -\begin{equation*} -\frac{d\langle E_L[\alpha]\rangle}{d\alpha} = \alpha-\frac{1}{4\alpha^3} -\end{equation*} -$$ - -and a second derivative which is always positive (meaning that we find a minimum) -$$ -\begin{equation*} -\hat{A}= \frac{d^2\langle E_L[\alpha]\rangle}{d\alpha^2} = 1+\frac{3}{4\alpha^4} -\end{equation*} -$$ -
- - -
-We get then -$$ -\begin{equation*} -\alpha_{i+1}=\frac{4}{3}\alpha_i-\frac{\alpha_i^4}{3\alpha_{i+1}^3}, -\end{equation*} -$$ - -which can be rewritten as -$$ -\begin{equation*} -\alpha_{i+1}^4-\frac{4}{3}\alpha_i\alpha_{i+1}^4+\frac{1}{3}\alpha_i^4. -\end{equation*} -$$ -
- - -
-In the CG method we define so-called conjugate directions and two vectors -\( \hat{s} \) and \( \hat{t} \) -are said to be -conjugate if -$$ -\begin{equation*} -\hat{s}^T\hat{A}\hat{t}= 0. -\end{equation*} -$$ - -The philosophy of the CG method is to perform searches in various conjugate directions -of our vectors \( \hat{x}_i \) obeying the above criterion, namely -$$ -\begin{equation*} -\hat{x}_i^T\hat{A}\hat{x}_j= 0. -\end{equation*} -$$ - -Two vectors are conjugate if they are orthogonal with respect to -this inner product. Being conjugate is a symmetric relation: if \( \hat{s} \) is conjugate to \( \hat{t} \), then \( \hat{t} \) is conjugate to \( \hat{s} \). -
- - -
-An example is given by the eigenvectors of the matrix -$$ -\begin{equation*} -\hat{v}_i^T\hat{A}\hat{v}_j= \lambda\hat{v}_i^T\hat{v}_j, -\end{equation*} -$$ - -which is zero unless \( i=j \). -
- - -
-Assume now that we have a symmetric positive-definite matrix \( \hat{A} \) of size -\( n\times n \). At each iteration \( i+1 \) we obtain the conjugate direction of a vector -$$ -\begin{equation*} -\hat{x}_{i+1}=\hat{x}_{i}+\alpha_i\hat{p}_{i}. -\end{equation*} -$$ - -We assume that \( \hat{p}_{i} \) is a sequence of \( n \) mutually conjugate directions. -Then the \( \hat{p}_{i} \) form a basis of \( R^n \) and we can expand the solution -$ \hat{A}\hat{x} = \hat{b}$ in this basis, namely - -$$ -\begin{equation*} - \hat{x} = \sum^{n}_{i=1} \alpha_i \hat{p}_i. -\end{equation*} -$$ -
- - -
-The coefficients are given by -$$ -\begin{equation*} - \mathbf{A}\mathbf{x} = \sum^{n}_{i=1} \alpha_i \mathbf{A} \mathbf{p}_i = \mathbf{b}. -\end{equation*} -$$ - -Multiplying with \( \hat{p}_k^T \) from the left gives - -$$ -\begin{equation*} - \hat{p}_k^T \hat{A}\hat{x} = \sum^{n}_{i=1} \alpha_i\hat{p}_k^T \hat{A}\hat{p}_i= \hat{p}_k^T \hat{b}, -\end{equation*} -$$ - -and we can define the coefficients \( \alpha_k \) as - -$$ -\begin{equation*} - \alpha_k = \frac{\hat{p}_k^T \hat{b}}{\hat{p}_k^T \hat{A} \hat{p}_k} -\end{equation*} -$$ -
- - -
- -
-If we choose the conjugate vectors \( \hat{p}_k \) carefully, -then we may not need all of them to obtain a good approximation to the solution -\( \hat{x} \). -We want to regard the conjugate gradient method as an iterative method. -This will us to solve systems where \( n \) is so large that the direct -method would take too much time. - -
-We denote the initial guess for \( \hat{x} \) as \( \hat{x}_0 \). -We can assume without loss of generality that -$$ -\begin{equation*} -\hat{x}_0=0, -\end{equation*} -$$ - -or consider the system -$$ -\begin{equation*} -\hat{A}\hat{z} = \hat{b}-\hat{A}\hat{x}_0, -\end{equation*} -$$ - -instead. -
- - -
-One can show that the solution \( \hat{x} \) is also the unique minimizer of the quadratic form -$$ -\begin{equation*} - f(\hat{x}) = \frac{1}{2}\hat{x}^T\hat{A}\hat{x} - \hat{x}^T \hat{x} , \quad \hat{x}\in\mathbf{R}^n. -\end{equation*} -$$ - -This suggests taking the first basis vector \( \hat{p}_1 \) -to be the gradient of \( f \) at \( \hat{x}=\hat{x}_0 \), -which equals -$$ -\begin{equation*} -\hat{A}\hat{x}_0-\hat{b}, -\end{equation*} -$$ - -and -\( \hat{x}_0=0 \) it is equal \( -\hat{b} \). -The other vectors in the basis will be conjugate to the gradient, -hence the name conjugate gradient method. -
- - -
-Let \( \hat{r}_k \) be the residual at the \( k \)-th step: -$$ -\begin{equation*} -\hat{r}_k=\hat{b}-\hat{A}\hat{x}_k. -\end{equation*} -$$ - -Note that \( \hat{r}_k \) is the negative gradient of \( f \) at -\( \hat{x}=\hat{x}_k \), -so the gradient descent method would be to move in the direction \( \hat{r}_k \). -Here, we insist that the directions \( \hat{p}_k \) are conjugate to each other, -so we take the direction closest to the gradient \( \hat{r}_k \) -under the conjugacy constraint. -This gives the following expression -$$ -\begin{equation*} -\hat{p}_{k+1}=\hat{r}_k-\frac{\hat{p}_k^T \hat{A}\hat{r}_k}{\hat{p}_k^T\hat{A}\hat{p}_k} \hat{p}_k. -\end{equation*} -$$ -
- - -
-We can also compute the residual iteratively as -$$ -\begin{equation*} -\hat{r}_{k+1}=\hat{b}-\hat{A}\hat{x}_{k+1}, - \end{equation*} -$$ - -which equals -$$ -\begin{equation*} -\hat{b}-\hat{A}(\hat{x}_k+\alpha_k\hat{p}_k), - \end{equation*} -$$ - -or -$$ -\begin{equation*} -(\hat{b}-\hat{A}\hat{x}_k)-\alpha_k\hat{A}\hat{p}_k, - \end{equation*} -$$ - -which gives - -$$ -\begin{equation*} -\hat{r}_{k+1}=\hat{r}_k-\hat{A}\hat{p}_{k}, - \end{equation*} -$$ -
- - -
# Importing various packages
-from math import exp, sqrt
from random import random, seed
import numpy as np
import matplotlib.pyplot as plt
@@ -890,11 +337,15 @@ plt.ylabel(r
plt.title(r'Random numbers ')
plt.show()
+ + +
# Importing various packages
-from math import exp, sqrt
from random import random, seed
import numpy as np
import matplotlib.pyplot as plt
@@ -910,6 +361,880 @@ sgdreg = SGDRegressor(n_iter .fit(x,y.ravel())
print(sgdreg.intercept_, sgdreg.coef_)
+ + +
+First we give the definition of a convex set: A set \( C \) in +\( \mathbb{R}^n \) is said to be convex if, for all \( x \) and \( y \) in \( C \) and +all \( t \in (0,1) \) , the point \( (1 − t)x + ty \) also belongs to +C. Geometrically this means that every point on the line segment +connecting \( x \) and \( y \) is in \( C \) as discussed below. + +
+The convex subsets of \( \mathbb{R} \) are the intervals of +\( \mathbb{R} \). Examples of convex sets of \( \mathbb{R}^2 \) are the +regular polygons (triangles, rectangles, pentagons, etc...). + +
+ + +
+Convex function: Let \( X \subset \mathbb{R}^n \) be a convex set. Assume that the function \( f: X \rightarrow \mathbb{R} \) is continuous, then \( f \) is said to be convex if $$f(tx_1 + (1-t)x_2) \leq tf(x_1) + (1-t)f(x_2) $$ for all \( x_1, x_2 \in X \) and for all \( t \in [0,1] \). If \( \leq \) is replaced with a strict inequaltiy in the definition, we demand \( x_1 \neq x_2 \) and \( t\in(0,1) \) then \( f \) is said to be strictly convex. For a single variable function, convexity means that if you draw a straight line connecting \( f(x_1) \) and \( f(x_2) \), the value of the function on the interval \( [x_1,x_2] \) is always below the line as illustrated below. + +
+ + +
+In the following we state first and second-order conditions which +ensures convexity of a function \( f \). We write \( D_f \) to denote the +domain of \( f \), i.e the subset of \( R^n \) where \( f \) is defined. For more +details and proofs we refer to: S. Boyd and L. Vandenberghe. Convex +Optimization. Cambridge University Press, http://stanford.edu/ +boyd/cvxbook/, 2004. + +
+
+Suppose \( f \) is differentiable (i.e \( \nabla f(x) \) is well defined for +all \( x \) in the domain of \( f \)). Then \( f \) is convex if and only if \( D_f \) +is a convex set and $$f(y) \geq f(x) + \nabla f(x)^T (y-x) $$ holds +for all \( x,y \in D_f \). This condition means that for a convex function +the first order Taylor expansion (right hand side above) at any point +a global under estimator of the function. To convince yourself you can +make a drawing of f(x) = x^2+1 and draw the tangent line to \( f(x) \) and +note that it is always below the graph. +
+
+Assume that \( f \) is twice +differentiable, i.e the Hessian matrix exists at each point in +\( D_f \). Then \( f \) is convex if and only if \( D_f \) is a convex set and its +Hessian is positive semi-definite for all \( x\in D_f \). For a +single-variable function this reduces to \( f''(x) \geq +0 \). Geometrically this means that \( f \) has nonnegative curvature +everywhere. +
+This condition is particularly useful since it gives us an procedure for determining if the function under consideration is convex, apart from using the definition. + +
+ + +
+The next result is of great importance to us and the reason why we are +going on about convex functions. In machine learning we frequently +have to minimize a loss/cost function in order to find the best +parameters for the model we are considering. Ideally we want the +global minimum, however for high-dimensional models it is hard to know +if we have local or global minimum. However, if the cost/loss function +is convex the following result provides invaluable information: + +
+
+Consider the problem of finding \( x \in \mathbb{R}^n \) such that \( f(x) \) +is minimal, where \( f \) is convex and differentiable. Then, any point +\( x^* \) that satisfies \( \nabla f(x^*) = 0 \) is a global minimum. +
+ + +
+ + +
+We will use linear regression as a case study for the gradient descent methods. Linear regression is a great test case for the gradient descent methods discussed in the lectures since it has several desirable properties such as: + +
+ + +
+Let \( \mathbf{y} = (y_1,\cdots,y_n)^T \), \( \mathbf{\hat{y}} = (\hat{y}_1,\cdots,\hat{y}_n)^T \) and \( \theta = (\theta_0, \theta_1)^T \) + +
+t is convenient to write \( \mathbf{\hat{y}} = X\theta \) where \( X \in \mathbb{R}^{100 \times 2} \) is the design matrix given by +$$ +\begin{equation} +X \equiv \begin{bmatrix} +1 & x_1 \\ +\vdots & \vdots \\ +1 & x_{100} & \\ +\end{bmatrix}. +\label{_auto1} +\end{equation} +$$ + +The loss function is given by +$$ +C(\theta) = ||X\theta-\mathbf{y}||^2 = ||X\theta||^2 - 2 \mathbf{y}^T X\theta + ||\mathbf{y}||^2 = \sum_{i=1}^{100} (\theta_0 + \theta_1 x_i)^2 - 2 y_i (\theta_0 + \theta_1 x_i) + y_i^2 +$$ + +and we want to find \( \theta \) such that \( C(\theta) \) is minimized. + +
+ + +
+Computing \( \partial C(\theta) / \partial \theta_0 \) and \( \partial C(\theta) / \partial \theta_1 \) we can show that the gradient can be written as +$$ +\nabla_\theta C(\theta) = (\partial C(\theta) / \partial \theta_0, \partial C(\theta) / \partial \theta_1)^T = 2\begin{bmatrix} \sum_{i=1}^{100} \left(\theta_0+\theta_1x_i-y_i\right) \\ +\sum_{i=1}^{100}\left( x_i (\theta_0+\theta_1x_i)-y_ix_i\right) \\ +\end{bmatrix} = 2X^T(X\theta - \mathbf{y}), +$$ + +where \( X \) is the design matrix defined above. + +
+ + +
+ + +
+We can now write a program that minimizes \( C(\theta) \) using the gradient descent method with a constant learning rate \( \gamma \) according to +$$ +\theta_{k+1} = \theta_k - \gamma \nabla_\theta C(\theta_k), \ k=0,1,\cdots +$$ + +
+We can use the expression we computed for the gradient and let use a +\( \theta_0 \) be chosen randomly and let \( \gamma = 0.001 \). Stop iterating +when \( ||\nabla_\theta C(\theta_k) || < \epsilon = 10^{-8} \). + +
+And finally we can compare our solution for \( \theta \) with the analytic result given by +\( \theta= (X^TX)^{-1} X^T \mathbf{y} \). +
+ + +
import numpy as np
+
+"""
+The following setup is just a suggestion, feel free to write it the way you like.
+"""
+
+#Setup problem described in the exercise
+N = 100 #Nr of datapoints
+M = 2 #Nr of features
+x = np.random.rand(N) #Uniformly generated x-values in [0,1]
+y = 5*x**2 + 0.1*np.random.randn(N)
+X = np.c_[np.ones(N),x] #Construct design matrix
+
+#Compute theta according to normal equations to compare with GD solution
+Xt_X_inv = np.linalg.inv(np.dot(X.T,X))
+Xt_y = np.dot(X.transpose(),y)
+theta_NE = np.dot(Xt_X_inv,Xt_y)
+print(theta_NE)
++ + +
+We have also discussed Ridge regression where the loss function contains a regularized given by the \( L_2 \) norm of \( \theta \), +$$ +C_{\text{ridge}}(\theta) = ||X\theta -\mathbf{y}||^2 + \lambda ||\theta||^2, \ \lambda \geq 0. +$$ + +
+In order to minimize \( C_{\text{ridge}}(\theta) \) using GD we only have adjust the gradient as follows +$$ +\nabla_\theta C_{\text{ridge}}(\theta) = 2\begin{bmatrix} \sum_{i=1}^{100} \left(\theta_0+\theta_1x_i-y_i\right) \\ +\sum_{i=1}^{100}\left( x_i (\theta_0+\theta_1x_i)-y_ix_i\right) \\ +\end{bmatrix} + 2\lambda\begin{bmatrix} \theta_0 \\ \theta_1\end{bmatrix} = 2 (X^T(X\theta - \mathbf{y})+\lambda \theta). +$$ + +
+We can now extend our program to minimize \( C_{\text{ridge}}(\theta) \) using gradient descent and compare with the analytical solution given by +$$ +\theta_{\text{ridge}} = \left(X^T X + \lambda I_{2 \times 2} \right)^{-1} X^T \mathbf{y}, +$$ + +for \( \lambda = {0,1,10,50,100} \) (\( \lambda = 0 \) corresponds to ordinary least squares). +We can then compute \( ||\theta_{\text{ridge}}|| \) for each \( \lambda \). + +
+ + +
import numpy as np
+
+"""
+The following setup is just a suggestion, feel free to write it the way you like.
+"""
+
+#Setup problem described in the exercise
+N = 100 #Nr of datapoints
+M = 2 #Nr of features
+x = np.random.rand(N)
+y = 5*x**2 + 0.1*np.random.randn(N)
+
+
+#Compute analytic theta for Ridge regression
+X = np.c_[np.ones(N),x]
+XT_X = np.dot(X.T,X)
+
+l = 0.1 #Ridge parameter lambda
+Id = np.eye(XT_X.shape[0])
+
+Z = np.linalg.inv(XT_X+l*Id)
+theta_ridge = np.dot(Z,np.dot(X.T,y))
+
+print(theta_ridge)
+print(np.linalg.norm(theta_ridge)) #||theta||
++ + +
+Stochastic gradient descent (SGD) and variants thereof address some of +the shortcomings of the Gradient descent method discussed above. + +
+The underlying idea of SGD comes from the observation that the cost +function, which we want to minimize, can almost always be written as a +sum over \( n \) datapoints \( \{\mathbf{x}_i\}_{i=1}^n \), +$$ +C(\mathbf{\theta}) = \sum_{i=1}^n c_i(\mathbf{x}_i, +\mathbf{\theta}). +$$ + +
+ + +
+This in turn means that the gradient can be +computed as a sum over \( i \)-gradients +$$ +\nabla_\theta C(\mathbf{\theta}) = \sum_i^n \nabla_\theta c_i(\mathbf{x}_i, +\mathbf{\theta}). +$$ + +
+Stochasticity/randomness is introduced by only taking the +gradient on a subset of the data called minibatches. If there are \( n \) +datapoints and the size of each minibatch is \( M \), there will be \( n/M \) +minibatches. We denote these minibatches by \( B_k \) where +\( k=1,\cdots,n/M \). + +
+ + +
+The idea is now to approximate the gradient by replacing the sum over +all datapoints with a sum over the datapoints in one the minibatches +picked at random in each gradient descent step +$$ +\nabla_\theta +C(\mathbf{\theta}) = \sum_{i=1}^n \nabla_\theta c_i(\mathbf{x}_i, +\mathbf{\theta}) \rightarrow \sum_{i \in B_k}^n \nabla_\theta +c_i(\mathbf{x}_i, \mathbf{\theta}). +$$ + +
+ + +
+Thus a gradient descent step now looks like +$$ +\theta_{j+1} = \theta_j - \gamma_j \sum_{i \in B_k}^n \nabla_\theta c_i(\mathbf{x}_i, +\mathbf{\theta}) +$$ + +
+where \( k \) is picked at random with equal +probability from \( [1,n/M] \). An iteration over the number of +minibathces (n/M) is commonly referred to as an epoch. Thus it is +typical to choose a number of epochs and for each epoch iterate over +the number of minibatches, as exemplified in the code below. + +
+ + +
+ + +
import numpy as np
+
+n = 100 #100 datapoints
+M = 5 #size of each minibatch
+m = int(n/M) #number of minibatches
+n_epochs = 10 #number of epochs
+
+j = 0
+for epoch in range(1,n_epochs+1):
+ for i in range(m):
+ k = np.random.randint(m) #Pick the k-th minibatch at random
+ #Compute the gradient using the data in minibatch Bk
+ #Compute new suggestion for theta
+ j += 1
++Taking the gradient only on a subset of the data has two important +benefits. First, it introduces randomness which decreases the chance +that our opmization scheme gets stuck in a local minima. Second, if +the size of the minibatches are small relative to the number of +datapoints (\( M < n \)), the computation of the gradient is much +cheaper since we sum over the datapoints in the k-th minibatch and not +all \( n \) datapoints. + +
+ + +
+A natural question is when do we stop the search for a new minimum? +One possibility is to compute the full gradient after a given number +of epochs and check if the norm of the gradient is smaller than some +threshold and stop if true. However, the condition that the gradient +is zero is valid also for local minima, so this would only tell us +that we are close to a local/global minimum. However, we could also +evaluate the cost function at this point, store the result and +continue the search. If the test kicks in at a later stage we can +compare the values of the cost function and keep the \( \theta \) that +gave the lowest value. + +
+ + +
+Another approach is to let the step length \( \gamma_j \) depend on the +number of epochs in such a way that it becomes very small after a +reasonable time such that we do not move at all. + +
+As an example, let \( e = 0,1,2,3,\cdots \) denote the current epoch and let \( t_0, t_1 > 0 \) be two fixed numbers. Furthermore, let \( t = e \cdot m + i \) where \( m \) is the number of minibatches and \( i=0,\cdots,m-1 \). Then the function $$\gamma_j(t; t_0, t_1) = \frac{t_0}{t+t_1} $$ goes to zero as the number of epochs gets large. I.e. we start with a step length \( \gamma_j (0; t_0, t_1) = t_0/t_1 \) which decays in time \( t \). + +
+In this way we can fix the number of epochs, compute \( \theta \) and +evaluate the cost function at the end. Repeating the computation will +give a different result since the scheme is random by design. Then we +pick the final \( \theta \) that gives the lowest value of the cost +function. + +
+ + +
import numpy as np
+
+def step_length(t,t0,t1):
+ return t0/(t+t1)
+
+n = 100 #100 datapoints
+M = 5 #size of each minibatch
+m = int(n/M) #number of minibatches
+n_epochs = 500 #number of epochs
+t0 = 1.0
+t1 = 10
+
+gamma_j = t0/t1
+j = 0
+for epoch in range(1,n_epochs+1):
+ for i in range(m):
+ k = np.random.randint(m) #Pick the k-th minibatch at random
+ #Compute the gradient using the data in minibatch Bk
+ #Compute new suggestion for theta
+ t = epoch*m+i
+ gamma_j = step_length(t,t0,t1)
+ j += 1
+
+print("gamma_j after %d epochs: %g" % (n_epochs,gamma_j))
++ + +
+The success of the CG method for finding solutions of non-linear problems is based +on the theory of conjugate gradients for linear systems of equations. It belongs +to the class of iterative methods for solving problems from linear algebra of the type +$$ +\begin{equation*} + \hat{A}\hat{x} = \hat{b}. +\end{equation*} +$$ + +In the iterative process we end up with a problem like + +$$ +\begin{equation*} + \hat{r}= \hat{b}-\hat{A}\hat{x}, +\end{equation*} +$$ + +where \( \hat{r} \) is the so-called residual or error in the iterative process. + +
+When we have found the exact solution, \( \hat{r}=0 \). +
+ + +
+ +
+The residual is zero when we reach the minimum of the quadratic equation +$$ +\begin{equation*} + P(\hat{x})=\frac{1}{2}\hat{x}^T\hat{A}\hat{x} - \hat{x}^T\hat{b}, +\end{equation*} +$$ + +with the constraint that the matrix \( \hat{A} \) is positive definite and symmetric. +If we search for a minimum of the quantum mechanical variance, then the matrix +\( \hat{A} \), which is called the Hessian, is given by the second-derivative of the function we want to minimize. This quantity is always positive definite. In our case this corresponds normally to the second derivative of the energy. +
+ + +
+We seek the minimum of the energy or the variance as function of various variational parameters. +In our case we have thus a function \( f \) whose minimum we are seeking. +In Newton's method we set \( \nabla f = 0 \) and we can thus compute the next iteration point +$$ +\begin{equation*} +\hat{x}-\hat{x}_i=\hat{A}^{-1}\nabla f(\hat{x}_i). +\end{equation*} +$$ + +Subtracting this equation from that of \( \hat{x}_{i+1} \) we have +$$ +\begin{equation*} +\hat{x}_{i+1}-\hat{x}_i=\hat{A}^{-1}(\nabla f(\hat{x}_{i+1})-\nabla f(\hat{x}_i)). +\end{equation*} +$$ +
+ + +
+The function \( f \) can be either the energy or the variance. If we choose the energy then we have +$$ +\begin{equation*} +\hat{\alpha}_{i+1}-\hat{\alpha}_i=\hat{A}^{-1}(\nabla E(\hat{\alpha}_{i+1})-\nabla E(\hat{\alpha}_i)). +\end{equation*} +$$ + +In the simple harmonic oscillator model, the gradient and the Hessian \( \hat{A} \) are +$$ +\begin{equation*} +\frac{d\langle E_L[\alpha]\rangle}{d\alpha} = \alpha-\frac{1}{4\alpha^3} +\end{equation*} +$$ + +and a second derivative which is always positive (meaning that we find a minimum) +$$ +\begin{equation*} +\hat{A}= \frac{d^2\langle E_L[\alpha]\rangle}{d\alpha^2} = 1+\frac{3}{4\alpha^4} +\end{equation*} +$$ +
+ + +
+We get then +$$ +\begin{equation*} +\alpha_{i+1}=\frac{4}{3}\alpha_i-\frac{\alpha_i^4}{3\alpha_{i+1}^3}, +\end{equation*} +$$ + +which can be rewritten as +$$ +\begin{equation*} +\alpha_{i+1}^4-\frac{4}{3}\alpha_i\alpha_{i+1}^4+\frac{1}{3}\alpha_i^4. +\end{equation*} +$$ +
+ + +
+In the CG method we define so-called conjugate directions and two vectors +\( \hat{s} \) and \( \hat{t} \) +are said to be +conjugate if +$$ +\begin{equation*} +\hat{s}^T\hat{A}\hat{t}= 0. +\end{equation*} +$$ + +The philosophy of the CG method is to perform searches in various conjugate directions +of our vectors \( \hat{x}_i \) obeying the above criterion, namely +$$ +\begin{equation*} +\hat{x}_i^T\hat{A}\hat{x}_j= 0. +\end{equation*} +$$ + +Two vectors are conjugate if they are orthogonal with respect to +this inner product. Being conjugate is a symmetric relation: if \( \hat{s} \) is conjugate to \( \hat{t} \), then \( \hat{t} \) is conjugate to \( \hat{s} \). +
+ + +
+An example is given by the eigenvectors of the matrix +$$ +\begin{equation*} +\hat{v}_i^T\hat{A}\hat{v}_j= \lambda\hat{v}_i^T\hat{v}_j, +\end{equation*} +$$ + +which is zero unless \( i=j \). +
+ + +
+Assume now that we have a symmetric positive-definite matrix \( \hat{A} \) of size +\( n\times n \). At each iteration \( i+1 \) we obtain the conjugate direction of a vector +$$ +\begin{equation*} +\hat{x}_{i+1}=\hat{x}_{i}+\alpha_i\hat{p}_{i}. +\end{equation*} +$$ + +We assume that \( \hat{p}_{i} \) is a sequence of \( n \) mutually conjugate directions. +Then the \( \hat{p}_{i} \) form a basis of \( R^n \) and we can expand the solution +$ \hat{A}\hat{x} = \hat{b}$ in this basis, namely + +$$ +\begin{equation*} + \hat{x} = \sum^{n}_{i=1} \alpha_i \hat{p}_i. +\end{equation*} +$$ +
+ + +
+The coefficients are given by +$$ +\begin{equation*} + \mathbf{A}\mathbf{x} = \sum^{n}_{i=1} \alpha_i \mathbf{A} \mathbf{p}_i = \mathbf{b}. +\end{equation*} +$$ + +Multiplying with \( \hat{p}_k^T \) from the left gives + +$$ +\begin{equation*} + \hat{p}_k^T \hat{A}\hat{x} = \sum^{n}_{i=1} \alpha_i\hat{p}_k^T \hat{A}\hat{p}_i= \hat{p}_k^T \hat{b}, +\end{equation*} +$$ + +and we can define the coefficients \( \alpha_k \) as + +$$ +\begin{equation*} + \alpha_k = \frac{\hat{p}_k^T \hat{b}}{\hat{p}_k^T \hat{A} \hat{p}_k} +\end{equation*} +$$ +
+ + +
+ +
+If we choose the conjugate vectors \( \hat{p}_k \) carefully, +then we may not need all of them to obtain a good approximation to the solution +\( \hat{x} \). +We want to regard the conjugate gradient method as an iterative method. +This will us to solve systems where \( n \) is so large that the direct +method would take too much time. + +
+We denote the initial guess for \( \hat{x} \) as \( \hat{x}_0 \). +We can assume without loss of generality that +$$ +\begin{equation*} +\hat{x}_0=0, +\end{equation*} +$$ + +or consider the system +$$ +\begin{equation*} +\hat{A}\hat{z} = \hat{b}-\hat{A}\hat{x}_0, +\end{equation*} +$$ + +instead. +
+ + +
+One can show that the solution \( \hat{x} \) is also the unique minimizer of the quadratic form +$$ +\begin{equation*} + f(\hat{x}) = \frac{1}{2}\hat{x}^T\hat{A}\hat{x} - \hat{x}^T \hat{x} , \quad \hat{x}\in\mathbf{R}^n. +\end{equation*} +$$ + +This suggests taking the first basis vector \( \hat{p}_1 \) +to be the gradient of \( f \) at \( \hat{x}=\hat{x}_0 \), +which equals +$$ +\begin{equation*} +\hat{A}\hat{x}_0-\hat{b}, +\end{equation*} +$$ + +and +\( \hat{x}_0=0 \) it is equal \( -\hat{b} \). +The other vectors in the basis will be conjugate to the gradient, +hence the name conjugate gradient method. +
+ + +
+Let \( \hat{r}_k \) be the residual at the \( k \)-th step: +$$ +\begin{equation*} +\hat{r}_k=\hat{b}-\hat{A}\hat{x}_k. +\end{equation*} +$$ + +Note that \( \hat{r}_k \) is the negative gradient of \( f \) at +\( \hat{x}=\hat{x}_k \), +so the gradient descent method would be to move in the direction \( \hat{r}_k \). +Here, we insist that the directions \( \hat{p}_k \) are conjugate to each other, +so we take the direction closest to the gradient \( \hat{r}_k \) +under the conjugacy constraint. +This gives the following expression +$$ +\begin{equation*} +\hat{p}_{k+1}=\hat{r}_k-\frac{\hat{p}_k^T \hat{A}\hat{r}_k}{\hat{p}_k^T\hat{A}\hat{p}_k} \hat{p}_k. +\end{equation*} +$$ +
+ + +
+We can also compute the residual iteratively as +$$ +\begin{equation*} +\hat{r}_{k+1}=\hat{b}-\hat{A}\hat{x}_{k+1}, + \end{equation*} +$$ + +which equals +$$ +\begin{equation*} +\hat{b}-\hat{A}(\hat{x}_k+\alpha_k\hat{p}_k), + \end{equation*} +$$ + +or +$$ +\begin{equation*} +(\hat{b}-\hat{A}\hat{x}_k)-\alpha_k\hat{A}\hat{p}_k, + \end{equation*} +$$ + +which gives + +$$ +\begin{equation*} +\hat{r}_{k+1}=\hat{r}_k-\hat{A}\hat{p}_{k}, + \end{equation*} +$$ +
diff --git a/doc/pub/Splines/html/Splines-reveal.html b/doc/pub/Splines/html/Splines-reveal.html index 49b13bbd1..d9f121b49 100644 --- a/doc/pub/Splines/html/Splines-reveal.html +++ b/doc/pub/Splines/html/Splines-reveal.html @@ -3,9 +3,9 @@ - + -
@@ -148,7 +148,7 @@ MathJax.Hub.Config({
-
@@ -159,310 +159,703 @@ MathJax.Hub.Config({
-Cubic spline interpolation is among one of the most used
-methods for interpolating between data points where the arguments
-are organized as ascending series. In the library program we supply
-such a function, based on the so-called cubic spline method to be
-described below.
+
-A spline function consists of polynomial pieces defined on
-subintervals. The different subintervals are connected via
-various continuity relations.
-
-
-Assume we have at our disposal \( n+1 \) points \( x_0, x_1, \dots x_n \)
-arranged so that \( x_0 < x_1 < x_2 < \dots x_{n-1} < x_n \) (such points are called
-knots). A spline function \( s \) of degree \( k \) with \( n+1 \) knots is defined
-as follows
-
-
-As an example, consider a spline function of degree \( k=1 \) defined as follows
+The method of steepest descent The basic idea of gradient descent is
+that a function \( F(\mathbf{x}) \),
+\( \mathbf{x} \equiv (x_1,\cdots,x_n) \), decreases fastest if one goes from \( \bf {x} \) in the
+direction of the negative gradient \( -\nabla F(\mathbf{x}) \).
+
+
+It can be shown that if
+for \( \gamma_k \) small enough, then \( F(\mathbf{x}_{k+1}) \leq
+F(\mathbf{x}_k) \). This means that for a sufficiently small \( \gamma_k \)
+we are always moving towards smaller function values, i.e a minimum.
-The most commonly used spline function is the one with \( k=3 \), the so-called
-cubic spline function.
-Assume that we have in adddition to the \( n+1 \) knots a series of
-functions values \( y_0=f(x_0), y_1=f(x_1), \dots y_n=f(x_n) \).
-By definition, the polynomials \( s_{i-1} \) and \( s_i \)
-are thence supposed to interpolate the same point \( i \), that is
+The previous observation is the basis of the method of steepest
+descent, which is also referred to as just gradient descent (GD). One
+starts with an initial guess \( \mathbf{x}_0 \) for a minimum of \( F \) and
+compute new approximations according to
+
+The parameter \( \gamma_k \) is often referred to as the step length or
+the learning rate in the context of Machine Learning.
-Every subinterval provides in addition the \( 2n \) conditions
-
+In machine learing we are often faced with non-convex high dimensional cost functions with many local minimum. Since GD is deterministic we will get stuck in a local minimum, if the method converges, unless we have a very good intial guess. This also implies that the scheme is sensitive to the chosen initial condition.
-to be fulfilled. If we also assume that \( s' \) and \( s'' \) are continuous,
-then
-
+Note that the gradient is a function of \( \mathbf{x} =
+(x_1,\cdots,x_n) \) which makes it expensive to compute numerically.
-Using the last equation we define two values for the second derivative, namely
-
+Many of these shortcomings can be alleviated by introducing
+randomness. One such method is that of Stochastic Gradient Descent
+(SGD), see below
-Using the conditions \( s_i(x_i)=y_i \) and \( s_i(x_{i+1})=y_{i+1} \)
-we can in turn determine the constants \( c \) and \( d \) resulting in
+
+
+
+
+
+
+First we give the definition of a convex set: A set \( C \) in
+\( \mathbb{R}^n \) is said to be convex if, for all \( x \) and \( y \) in \( C \) and
+all \( t \in (0,1) \) , the point \( (1 − t)x + ty \) also belongs to
+C. Geometrically this means that every point on the line segment
+connecting \( x \) and \( y \) is in \( C \) as discussed below.
+
+
+The convex subsets of \( \mathbb{R} \) are the intervals of
+\( \mathbb{R} \). Examples of convex sets of \( \mathbb{R}^2 \) are the
+regular polygons (triangles, rectangles, pentagons, etc...).
+
+Convex function: Let \( X \subset \mathbb{R}^n \) be a convex set. Assume that the function \( f: X \rightarrow \mathbb{R} \) is continuous, then \( f \) is said to be convex if
+In the following we state first and second-order conditions which
+ensures convexity of a function \( f \). We write \( D_f \) to denote the
+domain of \( f \), i.e the subset of \( R^n \) where \( f \) is defined. For more
+details and proofs we refer to: S. Boyd and L. Vandenberghe. Convex
+Optimization. Cambridge University Press, http://stanford.edu/
+boyd/cvxbook/, 2004.
+
+
+
+Suppose \( f \) is differentiable (i.e \( \nabla f(x) \) is well defined for
+all \( x \) in the domain of \( f \)). Then \( f \) is convex if and only if \( D_f \)
+is a convex set and
+
+Assume that \( f \) is twice
+differentiable, i.e the Hessian matrix exists at each point in
+\( D_f \). Then \( f \) is convex if and only if \( D_f \) is a convex set and its
+Hessian is positive semi-definite for all \( x\in D_f \). For a
+single-variable function this reduces to \( f''(x) \geq
+0 \). Geometrically this means that \( f \) has nonnegative curvature
+everywhere.
+
+This condition is particularly useful since it gives us an procedure for determining if the function under consideration is convex, apart from using the definition.
+
+The next result is of great importance to us and the reason why we are
+going on about convex functions. In machine learning we frequently
+have to minimize a loss/cost function in order to find the best
+parameters for the model we are considering. Ideally we want the
+global minimum, however for high-dimensional models it is hard to know
+if we have local or global minimum. However, if the cost/loss function
+is convex the following result provides invaluable information:
+
+
+
+Consider the problem of finding \( x \in \mathbb{R}^n \) such that \( f(x) \)
+is minimal, where \( f \) is convex and differentiable. Then, any point
+\( x^* \) that satisfies \( \nabla f(x^*) = 0 \) is a global minimum.
+
+
+Hint: If you re-write the definition, \( f \) is convex if the following holds for all \( x,y \in D_f \) and any \( \lambda \in [0,1] \)
+
+\( f(x) = e^x \) is convex for \( x \in \mathbb{R} \).
+\( g(x) = -\ln(x) \) is convex for \( x \in (0,\infty) \).
+
+
+
+\( f(\alpha x) = |\alpha| f(x) \) for all \( \alpha \in \mathbb{R} \).
+\( f(x+y) \leq f(x) + f(y) \)
+\( f(x) \leq 0 \) for all \( x \in \mathbb{R}^n \) with equality if and only if \( x = 0 \)
+Using the definition of convexity, show that a function satisfying the properties above is convex (the third condition is not needed to show this).
+
+We will use linear regression as a case study for the gradient descent methods. Linear regression is a great test case for the gradient descent methods discussed in the lectures since it has several desirable properties such as:
+
+
+
+We revisit the example from homework set 1 where we had
+Let \( \mathbf{y} = (y_1,\cdots,y_n)^T \), \( \mathbf{\hat{y}} = (\hat{y}_1,\cdots,\hat{y}_n)^T \) and \( \theta = (\theta_0, \theta_1)^T \)
+
+
+t is convenient to write \( \mathbf{\hat{y}} = X\theta \) where \( X \in \mathbb{R}^{100 \times 2} \) is the design matrix given by
+
-How to determine the values of the second
-derivatives \( f_{i} \) and \( f_{i+1} \)? We use the continuity assumption
-of the first derivatives
+Computing \( \partial C(\theta) / \partial \theta_0 \) and \( \partial C(\theta) / \partial \theta_1 \) we can show that the gradient can be written as
-Gaussian elimination
+
-The functions supplied in the program library are spline and splint.
-In order to use cubic spline interpolation you need first to call
+We can now write a program that minimizes \( C(\theta) \) using the gradient descent method with a constant learning rate \( \gamma \) according to
+
+We can use the expression we computed for the gradient and let use a
+\( \theta_0 \) be chosen randomly and let \( \gamma = 0.001 \). Stop iterating
+when \( ||\nabla_\theta C(\theta_k) || < \epsilon = 10^{-8} \).
+
+
+And finally we can compare our solution for \( \theta \) with the analytic result given by
+\( \theta= (X^TX)^{-1} X^T \mathbf{y} \).
+
+
+
+
+We have also discussed Ridge regression where the loss function contains a regularized given by the \( L_2 \) norm of \( \theta \),
+
+In order to minimize \( C_{\text{ridge}}(\theta) \) using GD we only have adjust the gradient as follows
+
+We can now extend our program to minimize \( C_{\text{ridge}}(\theta) \) using gradient descent and compare with the analytical solution given by
+
-
-
+Stochastic gradient descent (SGD) and variants thereof address some of
+the shortcomings of the Gradient descent method discussed above.
+
+
+The underlying idea of SGD comes from the observation that the cost
+function, which we want to minimize, can almost always be written as a
+sum over \( n \) datapoints \( \{\mathbf{x}_i\}_{i=1}^n \),
+
+This in turn means that the gradient can be
+computed as a sum over \( i \)-gradients
+
+Stochasticity/randomness is introduced by only taking the
+gradient on a subset of the data called minibatches. If there are \( n \)
+datapoints and the size of each minibatch is \( M \), there will be \( n/M \)
+minibatches. We denote these minibatches by \( B_k \) where
+\( k=1,\cdots,n/M \).
+
+The idea is now to approximate the gradient by replacing the sum over
+all datapoints with a sum over the datapoints in one the minibatches
+picked at random in each gradient descent step
+
+Thus a gradient descent step now looks like
+
+where \( k \) is picked at random with equal
+probability from \( [1,n/M] \). An iteration over the number of
+minibathces (n/M) is commonly referred to as an epoch. Thus it is
+typical to choose a number of epochs and for each epoch iterate over
+the number of minibatches, as exemplified in the code below.
+
+
+
+
-This function takes as
-input \( x[0,..,n - 1] \) and \( y[0,..,n - 1] \) containing a tabulation
-\( y_i = f(x_i) \) with \( x_0 < x_1 < .. < x_{n - 1} \)
-together with the
-first derivatives of \( f(x) \) at \( x_0 \) and \( x_{n-1} \), respectively. Then the
-function returns \( y2[0,..,n-1] \) which contains the second derivatives of
-\( f(x_i) \) at each point \( x_i \). \( n \) is the number of points.
-This function provides the cubic spline interpolation for all subintervals
-and is called only once.
-
-Thereafter, if you wish to make various interpolations, you need to call the function
+A natural question is when do we stop the search for a new minimum?
+One possibility is to compute the full gradient after a given number
+of epochs and check if the norm of the gradient is smaller than some
+threshold and stop if true. However, the condition that the gradient
+is zero is valid also for local minima, so this would only tell us
+that we are close to a local/global minimum. However, we could also
+evaluate the cost function at this point, store the result and
+continue the search. If the test kicks in at a later stage we can
+compare the values of the cost function and keep the \( \theta \) that
+gave the lowest value.
+
+
+
+
+Another approach is to let the step length \( \gamma_j \) depend on the
+number of epochs in such a way that it becomes very small after a
+reasonable time such that we do not move at all.
+
+
+As an example, let \( e = 0,1,2,3,\cdots \) denote the current epoch and let \( t_0, t_1 > 0 \) be two fixed numbers. Furthermore, let \( t = e \cdot m + i \) where \( m \) is the number of minibatches and \( i=0,\cdots,m-1 \). Then the function
+In this way we can fix the number of epochs, compute \( \theta \) and
+evaluate the cost function at the end. Repeating the computation will
+give a different result since the scheme is random by design. Then we
+pick the final \( \theta \) that gives the lowest value of the cost
+function.
+
-
-
-which takes as input
-the tabulated values \( x[0,..,n - 1] \) and \( y[0,..,n - 1] \) and the output
-y2a[0,..,n - 1] from spline. It returns the value \( y \) corresponding
-to the point \( x \).
-
@@ -496,7 +889,7 @@ When we have found the exact solution, \( \hat{r}=0 \).
@@ -517,7 +910,7 @@ If we search for a minimum of the quantum mechanical variance, then the matrix
@@ -545,7 +938,7 @@ $$
@@ -580,7 +973,7 @@ $$
@@ -606,7 +999,7 @@ $$
@@ -639,7 +1032,7 @@ this inner product. Being conjugate is a symmetric relation: if \( \hat{s} \) is
@@ -658,7 +1051,7 @@ which is zero unless \( i=j \).
@@ -688,7 +1081,7 @@ $$
@@ -725,7 +1118,7 @@ $$
@@ -762,7 +1155,7 @@ instead.
@@ -795,7 +1188,7 @@ hence the name conjugate gradient method.
@@ -827,7 +1220,7 @@ $$
@@ -871,74 +1264,6 @@ $$
-
-
-
-
-
-
-
@@ -128,671 +149,106 @@ MathJax.Hub.Config({
-
-Cubic spline interpolation is among one of the most used
-methods for interpolating between data points where the arguments
-are organized as ascending series. In the library program we supply
-such a function, based on the so-called cubic spline method to be
-described below.
+
-A spline function consists of polynomial pieces defined on
-subintervals. The different subintervals are connected via
-various continuity relations.
-
-
-Assume we have at our disposal \( n+1 \) points \( x_0, x_1, \dots x_n \)
-arranged so that \( x_0 < x_1 < x_2 < \dots x_{n-1} < x_n \) (such points are called
-knots). A spline function \( s \) of degree \( k \) with \( n+1 \) knots is defined
-as follows
-
-
-As an example, consider a spline function of degree \( k=1 \) defined as follows
+The method of steepest descent The basic idea of gradient descent is
+that a function \( F(\mathbf{x}) \),
+\( \mathbf{x} \equiv (x_1,\cdots,x_n) \), decreases fastest if one goes from \( \bf {x} \) in the
+direction of the negative gradient \( -\nabla F(\mathbf{x}) \).
+
+
+It can be shown that if
$$
- s(x)=\begin{bmatrix} s_0(x)=a_0x+b_0 & x\in [x_0, x_1) \\
- s_1(x)=a_1x+b_1 & x\in [x_1, x_2) \\
- \dots & \dots \\
- s_{n-1}(x)=a_{n-1}x+b_{n-1} & x\in
- [x_{n-1}, x_n] \end{bmatrix}.
+\mathbf{x}_{k+1} = \mathbf{x}_k - \gamma_k \nabla F(\mathbf{x}_k), \ \ \gamma_k > 0
$$
-In this case the polynomial consists of series of straight lines
-connected to each other at every endpoint. The number of continuous
-derivatives is then \( k-1=0 \), as expected when we deal with straight lines.
-Such a polynomial is quite easy to construct given
-\( n+1 \) points \( x_0, x_1, \dots x_n \) and their corresponding
-function values.
-
+for \( \gamma_k \) small enough, then \( F(\mathbf{x}_{k+1}) \leq
+F(\mathbf{x}_k) \). This means that for a sufficiently small \( \gamma_k \)
+we are always moving towards smaller function values, i.e a minimum.
+
+
+
+
+The previous observation is the basis of the method of steepest
+descent, which is also referred to as just gradient descent (GD). One
+starts with an initial guess \( \mathbf{x}_0 \) for a minimum of \( F \) and
+compute new approximations according to
+
+$$
+\mathbf{x}_{k+1} = \mathbf{x}_k - \gamma_k \nabla F(\mathbf{x}_k), \ \ k \geq 0.
+$$
+
+
+The parameter \( \gamma_k \) is often referred to as the step length or
+the learning rate in the context of Machine Learning.
+
+
+
+
+
+Ideally the sequence \( \{ \mathbf{x}_k \}_{k=0} \) converges to a global minimum of the function \( F \). In general we do not know if we are in a global or local minimum. In the special case when \( F \) is a convex function, all local minima are also global minima, so in this case gradient descent can converge to the global solution. The advantage of this scheme is that it is conceptually simple and straightforward to implement. However the method in this form has some severe limitations:
+
+
+In machine learing we are often faced with non-convex high dimensional cost functions with many local minimum. Since GD is deterministic we will get stuck in a local minimum, if the method converges, unless we have a very good intial guess. This also implies that the scheme is sensitive to the chosen initial condition.
+
+
+Note that the gradient is a function of \( \mathbf{x} =
+(x_1,\cdots,x_n) \) which makes it expensive to compute numerically.
+
+
+
+
+
+GD is sensitive to the choice of learning rate \( \gamma_k \). This is due
+to the fact that we are only guaranteed that \( F(\mathbf{x}_{k+1}) \leq
+F(\mathbf{x}_k) \) for sufficiently small \( \gamma_k \). The problem is to
+determine an optimal learning rate. If the learning rate is chosen to
+small the method will take a long to converge and if it is to large we
+can experience erratic behavior.
+
+
+Many of these shortcomings can be alleviated by introducing
+randomness. One such method is that of Stochastic Gradient Descent
+(SGD), see below
-The most commonly used spline function is the one with \( k=3 \), the so-called
-cubic spline function.
-Assume that we have in adddition to the \( n+1 \) knots a series of
-functions values \( y_0=f(x_0), y_1=f(x_1), \dots y_n=f(x_n) \).
-By definition, the polynomials \( s_{i-1} \) and \( s_i \)
-are thence supposed to interpolate the same point \( i \), that is
-$$
- s_{i-1}(x_i)= y_i = s_i(x_i),
-$$
-
-with \( 1 \le i \le n-1 \). In total we have \( n \) polynomials of the
-type
-$$
- s_i(x)=a_{i0}+a_{i1}x+a_{i2}x^2+a_{i2}x^3,
-$$
-
-yielding \( 4n \) coefficients to determine.
-
-
-Every subinterval provides in addition the \( 2n \) conditions
-$$
- y_i = s(x_i),
-$$
-
-and
-$$
- s(x_{i+1})= y_{i+1},
-$$
-
-to be fulfilled. If we also assume that \( s' \) and \( s'' \) are continuous,
-then
-$$
- s'_{i-1}(x_i)= s'_i(x_i),
-$$
-
-yields \( n-1 \) conditions. Similarly,
-$$
- s''_{i-1}(x_i)= s''_i(x_i),
-$$
-
-results in additional \( n-1 \) conditions. In total we have \( 4n \) coefficients
-and \( 4n-2 \) equations to determine them, leaving us with \( 2 \) degrees of
-freedom to be determined.
-
-
-Using the last equation we define two values for the second derivative, namely
-$$
- s''_{i}(x_i)= f_i,
-$$
-
-and
-$$
- s''_{i}(x_{i+1})= f_{i+1},
-$$
-
-and setting up a straight line between \( f_i \) and \( f_{i+1} \) we have
-$$
- s_i''(x) = \frac{f_i}{x_{i+1}-x_i}(x_{i+1}-x)+
- \frac{f_{i+1}}{x_{i+1}-x_i}(x-x_i),
-$$
-
-and integrating twice one obtains
-$$
- s_i(x) = \frac{f_i}{6(x_{i+1}-x_i)}(x_{i+1}-x)^3+
- \frac{f_{i+1}}{6(x_{i+1}-x_i)}(x-x_i)^3
- +c(x-x_i)+d(x_{i+1}-x).
-$$
-
-
-Using the conditions \( s_i(x_i)=y_i \) and \( s_i(x_{i+1})=y_{i+1} \)
-we can in turn determine the constants \( c \) and \( d \) resulting in
-$$
-\begin{align}
- s_i(x) =&\frac{f_i}{6(x_{i+1}-x_i)}(x_{i+1}-x)^3+
- \frac{f_{i+1}}{6(x_{i+1}-x_i)}(x-x_i)^3 \nonumber \\
- +&(\frac{y_{i+1}}{x_{i+1}-x_i}-\frac{f_{i+1}(x_{i+1}-x_i)}{6})
- (x-x_i)+
- (\frac{y_{i}}{x_{i+1}-x_i}-\frac{f_{i}(x_{i+1}-x_i)}{6})
- (x_{i+1}-x).
-\label{_auto1}
-\end{align}
-$$
-
-
-How to determine the values of the second
-derivatives \( f_{i} \) and \( f_{i+1} \)? We use the continuity assumption
-of the first derivatives
-$$
- s'_{i-1}(x_i)= s'_i(x_i),
-$$
-
-and set \( x=x_i \). Defining \( h_i=x_{i+1}-x_i \) we obtain finally
-the following expression
-$$
- h_{i-1}f_{i-1}+2(h_{i}+h_{i-1})f_i+h_if_{i+1}=
- \frac{6}{h_i}(y_{i+1}-y_i)-\frac{6}{h_{i-1}}(y_{i}-y_{i-1}),
-$$
-
-and introducing the shorthands \( u_i=2(h_{i}+h_{i-1}) \),
-\( v_i=\frac{6}{h_i}(y_{i+1}-y_i)-\frac{6}{h_{i-1}}(y_{i}-y_{i-1}) \),
-we can reformulate the problem as a set of linear equations to be
-solved through e.g., Gaussian elemination
-
-
-Gaussian elimination
-$$
- \begin{bmatrix} u_1 & h_1 &0 &\dots & & & & \\
- h_1 & u_2 & h_2 &0 &\dots & & & \\
- 0 & h_2 & u_3 & h_3 &0 &\dots & & \\
- \dots& & \dots &\dots &\dots &\dots &\dots & \\
- &\dots & & &0 &h_{n-3} &u_{n-2} &h_{n-2} \\
- & && & &0 &h_{n-2} &u_{n-1} \end{bmatrix}
- \begin{bmatrix} f_1 \\
- f_2 \\
- f_3\\
- \dots \\
- f_{n-2} \\
- f_{n-1} \end{bmatrix} =
- \begin{bmatrix} v_1 \\
- v_2 \\
- v_3\\
- \dots \\
- v_{n-2}\\
- v_{n-1} \end{bmatrix}.
-$$
-
-Note that this is a set of tridiagonal equations and can be solved
-through only \( O(n) \) operations.
-
-
-The functions supplied in the program library are spline and splint.
-In order to use cubic spline interpolation you need first to call
-
-
-
-
-
-This function takes as
-input \( x[0,..,n - 1] \) and \( y[0,..,n - 1] \) containing a tabulation
-\( y_i = f(x_i) \) with \( x_0 < x_1 < .. < x_{n - 1} \)
-together with the
-first derivatives of \( f(x) \) at \( x_0 \) and \( x_{n-1} \), respectively. Then the
-function returns \( y2[0,..,n-1] \) which contains the second derivatives of
-\( f(x_i) \) at each point \( x_i \). \( n \) is the number of points.
-This function provides the cubic spline interpolation for all subintervals
-and is called only once.
-
-
-Thereafter, if you wish to make various interpolations, you need to call the function
-
-
-
-
-which takes as input
-the tabulated values \( x[0,..,n - 1] \) and \( y[0,..,n - 1] \) and the output
-y2a[0,..,n - 1] from spline. It returns the value \( y \) corresponding
-to the point \( x \).
-
-
-The success of the CG method for finding solutions of non-linear problems is based
-on the theory of conjugate gradients for linear systems of equations. It belongs
-to the class of iterative methods for solving problems from linear algebra of the type
-$$
-\begin{equation*}
- \hat{A}\hat{x} = \hat{b}.
-\end{equation*}
-$$
-
-In the iterative process we end up with a problem like
-
-$$
-\begin{equation*}
- \hat{r}= \hat{b}-\hat{A}\hat{x},
-\end{equation*}
-$$
-
-where \( \hat{r} \) is the so-called residual or error in the iterative process.
-
-
-When we have found the exact solution, \( \hat{r}=0 \).
-
-
-
-
-The residual is zero when we reach the minimum of the quadratic equation
-$$
-\begin{equation*}
- P(\hat{x})=\frac{1}{2}\hat{x}^T\hat{A}\hat{x} - \hat{x}^T\hat{b},
-\end{equation*}
-$$
-
-with the constraint that the matrix \( \hat{A} \) is positive definite and symmetric.
-If we search for a minimum of the quantum mechanical variance, then the matrix
-\( \hat{A} \), which is called the Hessian, is given by the second-derivative of the function we want to minimize. This quantity is always positive definite. In our case this corresponds normally to the second derivative of the energy.
-
-
-We seek the minimum of the energy or the variance as function of various variational parameters.
-In our case we have thus a function \( f \) whose minimum we are seeking.
-In Newton's method we set \( \nabla f = 0 \) and we can thus compute the next iteration point
-$$
-\begin{equation*}
-\hat{x}-\hat{x}_i=\hat{A}^{-1}\nabla f(\hat{x}_i).
-\end{equation*}
-$$
-
-Subtracting this equation from that of \( \hat{x}_{i+1} \) we have
-$$
-\begin{equation*}
-\hat{x}_{i+1}-\hat{x}_i=\hat{A}^{-1}(\nabla f(\hat{x}_{i+1})-\nabla f(\hat{x}_i)).
-\end{equation*}
-$$
-
-
-The function \( f \) can be either the energy or the variance. If we choose the energy then we have
-$$
-\begin{equation*}
-\hat{\alpha}_{i+1}-\hat{\alpha}_i=\hat{A}^{-1}(\nabla E(\hat{\alpha}_{i+1})-\nabla E(\hat{\alpha}_i)).
-\end{equation*}
-$$
-
-In the simple harmonic oscillator model, the gradient and the Hessian \( \hat{A} \) are
-$$
-\begin{equation*}
-\frac{d\langle E_L[\alpha]\rangle}{d\alpha} = \alpha-\frac{1}{4\alpha^3}
-\end{equation*}
-$$
-
-and a second derivative which is always positive (meaning that we find a minimum)
-$$
-\begin{equation*}
-\hat{A}= \frac{d^2\langle E_L[\alpha]\rangle}{d\alpha^2} = 1+\frac{3}{4\alpha^4}
-\end{equation*}
-$$
-
-
-We get then
-$$
-\begin{equation*}
-\alpha_{i+1}=\frac{4}{3}\alpha_i-\frac{\alpha_i^4}{3\alpha_{i+1}^3},
-\end{equation*}
-$$
-
-which can be rewritten as
-$$
-\begin{equation*}
-\alpha_{i+1}^4-\frac{4}{3}\alpha_i\alpha_{i+1}^4+\frac{1}{3}\alpha_i^4.
-\end{equation*}
-$$
-
-
-In the CG method we define so-called conjugate directions and two vectors
-\( \hat{s} \) and \( \hat{t} \)
-are said to be
-conjugate if
-$$
-\begin{equation*}
-\hat{s}^T\hat{A}\hat{t}= 0.
-\end{equation*}
-$$
-
-The philosophy of the CG method is to perform searches in various conjugate directions
-of our vectors \( \hat{x}_i \) obeying the above criterion, namely
-$$
-\begin{equation*}
-\hat{x}_i^T\hat{A}\hat{x}_j= 0.
-\end{equation*}
-$$
-
-Two vectors are conjugate if they are orthogonal with respect to
-this inner product. Being conjugate is a symmetric relation: if \( \hat{s} \) is conjugate to \( \hat{t} \), then \( \hat{t} \) is conjugate to \( \hat{s} \).
-
-
-An example is given by the eigenvectors of the matrix
-$$
-\begin{equation*}
-\hat{v}_i^T\hat{A}\hat{v}_j= \lambda\hat{v}_i^T\hat{v}_j,
-\end{equation*}
-$$
-
-which is zero unless \( i=j \).
-
-
-Assume now that we have a symmetric positive-definite matrix \( \hat{A} \) of size
-\( n\times n \). At each iteration \( i+1 \) we obtain the conjugate direction of a vector
-$$
-\begin{equation*}
-\hat{x}_{i+1}=\hat{x}_{i}+\alpha_i\hat{p}_{i}.
-\end{equation*}
-$$
-
-We assume that \( \hat{p}_{i} \) is a sequence of \( n \) mutually conjugate directions.
-Then the \( \hat{p}_{i} \) form a basis of \( R^n \) and we can expand the solution
-$ \hat{A}\hat{x} = \hat{b}$ in this basis, namely
-
-$$
-\begin{equation*}
- \hat{x} = \sum^{n}_{i=1} \alpha_i \hat{p}_i.
-\end{equation*}
-$$
-
-
-The coefficients are given by
-$$
-\begin{equation*}
- \mathbf{A}\mathbf{x} = \sum^{n}_{i=1} \alpha_i \mathbf{A} \mathbf{p}_i = \mathbf{b}.
-\end{equation*}
-$$
-
-Multiplying with \( \hat{p}_k^T \) from the left gives
-
-$$
-\begin{equation*}
- \hat{p}_k^T \hat{A}\hat{x} = \sum^{n}_{i=1} \alpha_i\hat{p}_k^T \hat{A}\hat{p}_i= \hat{p}_k^T \hat{b},
-\end{equation*}
-$$
-
-and we can define the coefficients \( \alpha_k \) as
-
-$$
-\begin{equation*}
- \alpha_k = \frac{\hat{p}_k^T \hat{b}}{\hat{p}_k^T \hat{A} \hat{p}_k}
-\end{equation*}
-$$
-
-
-
-
-If we choose the conjugate vectors \( \hat{p}_k \) carefully,
-then we may not need all of them to obtain a good approximation to the solution
-\( \hat{x} \).
-We want to regard the conjugate gradient method as an iterative method.
-This will us to solve systems where \( n \) is so large that the direct
-method would take too much time.
-
-
-We denote the initial guess for \( \hat{x} \) as \( \hat{x}_0 \).
-We can assume without loss of generality that
-$$
-\begin{equation*}
-\hat{x}_0=0,
-\end{equation*}
-$$
-
-or consider the system
-$$
-\begin{equation*}
-\hat{A}\hat{z} = \hat{b}-\hat{A}\hat{x}_0,
-\end{equation*}
-$$
-
-instead.
-
-
-One can show that the solution \( \hat{x} \) is also the unique minimizer of the quadratic form
-$$
-\begin{equation*}
- f(\hat{x}) = \frac{1}{2}\hat{x}^T\hat{A}\hat{x} - \hat{x}^T \hat{x} , \quad \hat{x}\in\mathbf{R}^n.
-\end{equation*}
-$$
-
-This suggests taking the first basis vector \( \hat{p}_1 \)
-to be the gradient of \( f \) at \( \hat{x}=\hat{x}_0 \),
-which equals
-$$
-\begin{equation*}
-\hat{A}\hat{x}_0-\hat{b},
-\end{equation*}
-$$
-
-and
-\( \hat{x}_0=0 \) it is equal \( -\hat{b} \).
-The other vectors in the basis will be conjugate to the gradient,
-hence the name conjugate gradient method.
-
-
-Let \( \hat{r}_k \) be the residual at the \( k \)-th step:
-$$
-\begin{equation*}
-\hat{r}_k=\hat{b}-\hat{A}\hat{x}_k.
-\end{equation*}
-$$
-
-Note that \( \hat{r}_k \) is the negative gradient of \( f \) at
-\( \hat{x}=\hat{x}_k \),
-so the gradient descent method would be to move in the direction \( \hat{r}_k \).
-Here, we insist that the directions \( \hat{p}_k \) are conjugate to each other,
-so we take the direction closest to the gradient \( \hat{r}_k \)
-under the conjugacy constraint.
-This gives the following expression
-$$
-\begin{equation*}
-\hat{p}_{k+1}=\hat{r}_k-\frac{\hat{p}_k^T \hat{A}\hat{r}_k}{\hat{p}_k^T\hat{A}\hat{p}_k} \hat{p}_k.
-\end{equation*}
-$$
-
-
-We can also compute the residual iteratively as
-$$
-\begin{equation*}
-\hat{r}_{k+1}=\hat{b}-\hat{A}\hat{x}_{k+1},
- \end{equation*}
-$$
-
-which equals
-$$
-\begin{equation*}
-\hat{b}-\hat{A}(\hat{x}_k+\alpha_k\hat{p}_k),
- \end{equation*}
-$$
-
-or
-$$
-\begin{equation*}
-(\hat{b}-\hat{A}\hat{x}_k)-\alpha_k\hat{A}\hat{p}_k,
- \end{equation*}
-$$
-
-which gives
-
-$$
-\begin{equation*}
-\hat{r}_{k+1}=\hat{r}_k-\hat{A}\hat{p}_{k},
- \end{equation*}
-$$
-
-
+
+
+
+
+First we give the definition of a convex set: A set \( C \) in
+\( \mathbb{R}^n \) is said to be convex if, for all \( x \) and \( y \) in \( C \) and
+all \( t \in (0,1) \) , the point \( (1 − t)x + ty \) also belongs to
+C. Geometrically this means that every point on the line segment
+connecting \( x \) and \( y \) is in \( C \) as discussed below.
+
+
+The convex subsets of \( \mathbb{R} \) are the intervals of
+\( \mathbb{R} \). Examples of convex sets of \( \mathbb{R}^2 \) are the
+regular polygons (triangles, rectangles, pentagons, etc...).
+
+
+
+Convex function: Let \( X \subset \mathbb{R}^n \) be a convex set. Assume that the function \( f: X \rightarrow \mathbb{R} \) is continuous, then \( f \) is said to be convex if $$f(tx_1 + (1-t)x_2) \leq tf(x_1) + (1-t)f(x_2) $$ for all \( x_1, x_2 \in X \) and for all \( t \in [0,1] \). If \( \leq \) is replaced with a strict inequaltiy in the definition, we demand \( x_1 \neq x_2 \) and \( t\in(0,1) \) then \( f \) is said to be strictly convex. For a single variable function, convexity means that if you draw a straight line connecting \( f(x_1) \) and \( f(x_2) \), the value of the function on the interval \( [x_1,x_2] \) is always below the line as illustrated below.
+
+
+
+In the following we state first and second-order conditions which
+ensures convexity of a function \( f \). We write \( D_f \) to denote the
+domain of \( f \), i.e the subset of \( R^n \) where \( f \) is defined. For more
+details and proofs we refer to: S. Boyd and L. Vandenberghe. Convex
+Optimization. Cambridge University Press, http://stanford.edu/
+boyd/cvxbook/, 2004.
+
+
+
+Suppose \( f \) is differentiable (i.e \( \nabla f(x) \) is well defined for
+all \( x \) in the domain of \( f \)). Then \( f \) is convex if and only if \( D_f \)
+is a convex set and $$f(y) \geq f(x) + \nabla f(x)^T (y-x) $$ holds
+for all \( x,y \in D_f \). This condition means that for a convex function
+the first order Taylor expansion (right hand side above) at any point
+a global under estimator of the function. To convince yourself you can
+make a drawing of f(x) = x^2+1 and draw the tangent line to \( f(x) \) and
+note that it is always below the graph.
+
+
+Assume that \( f \) is twice
+differentiable, i.e the Hessian matrix exists at each point in
+\( D_f \). Then \( f \) is convex if and only if \( D_f \) is a convex set and its
+Hessian is positive semi-definite for all \( x\in D_f \). For a
+single-variable function this reduces to \( f''(x) \geq
+0 \). Geometrically this means that \( f \) has nonnegative curvature
+everywhere.
+
+This condition is particularly useful since it gives us an procedure for determining if the function under consideration is convex, apart from using the definition.
+
+
+
+The next result is of great importance to us and the reason why we are
+going on about convex functions. In machine learning we frequently
+have to minimize a loss/cost function in order to find the best
+parameters for the model we are considering. Ideally we want the
+global minimum, however for high-dimensional models it is hard to know
+if we have local or global minimum. However, if the cost/loss function
+is convex the following result provides invaluable information:
+
+
+
+Consider the problem of finding \( x \in \mathbb{R}^n \) such that \( f(x) \)
+is minimal, where \( f \) is convex and differentiable. Then, any point
+\( x^* \) that satisfies \( \nabla f(x^*) = 0 \) is a global minimum.
+
+
+
+
+
+We will use linear regression as a case study for the gradient descent methods. Linear regression is a great test case for the gradient descent methods discussed in the lectures since it has several desirable properties such as:
+
+
+
+
+
+Let \( \mathbf{y} = (y_1,\cdots,y_n)^T \), \( \mathbf{\hat{y}} = (\hat{y}_1,\cdots,\hat{y}_n)^T \) and \( \theta = (\theta_0, \theta_1)^T \)
+
+
+t is convenient to write \( \mathbf{\hat{y}} = X\theta \) where \( X \in \mathbb{R}^{100 \times 2} \) is the design matrix given by
+$$
+\begin{equation}
+X \equiv \begin{bmatrix}
+1 & x_1 \\
+\vdots & \vdots \\
+1 & x_{100} & \\
+\end{bmatrix}.
+\label{_auto1}
+\end{equation}
+$$
+
+The loss function is given by
+$$
+C(\theta) = ||X\theta-\mathbf{y}||^2 = ||X\theta||^2 - 2 \mathbf{y}^T X\theta + ||\mathbf{y}||^2 = \sum_{i=1}^{100} (\theta_0 + \theta_1 x_i)^2 - 2 y_i (\theta_0 + \theta_1 x_i) + y_i^2
+$$
+
+and we want to find \( \theta \) such that \( C(\theta) \) is minimized.
+
+
+
+Computing \( \partial C(\theta) / \partial \theta_0 \) and \( \partial C(\theta) / \partial \theta_1 \) we can show that the gradient can be written as
+$$
+\nabla_\theta C(\theta) = (\partial C(\theta) / \partial \theta_0, \partial C(\theta) / \partial \theta_1)^T = 2\begin{bmatrix} \sum_{i=1}^{100} \left(\theta_0+\theta_1x_i-y_i\right) \\
+\sum_{i=1}^{100}\left( x_i (\theta_0+\theta_1x_i)-y_ix_i\right) \\
+\end{bmatrix} = 2X^T(X\theta - \mathbf{y}),
+$$
+
+where \( X \) is the design matrix defined above.
+
+
+
+
+We can now write a program that minimizes \( C(\theta) \) using the gradient descent method with a constant learning rate \( \gamma \) according to
+$$
+\theta_{k+1} = \theta_k - \gamma \nabla_\theta C(\theta_k), \ k=0,1,\cdots
+$$
+
+
+We can use the expression we computed for the gradient and let use a
+\( \theta_0 \) be chosen randomly and let \( \gamma = 0.001 \). Stop iterating
+when \( ||\nabla_\theta C(\theta_k) || < \epsilon = 10^{-8} \).
+
+
+And finally we can compare our solution for \( \theta \) with the analytic result given by
+\( \theta= (X^TX)^{-1} X^T \mathbf{y} \).
+
+
+
+
+
+
+
+We have also discussed Ridge regression where the loss function contains a regularized given by the \( L_2 \) norm of \( \theta \),
+$$
+C_{\text{ridge}}(\theta) = ||X\theta -\mathbf{y}||^2 + \lambda ||\theta||^2, \ \lambda \geq 0.
+$$
+
+
+In order to minimize \( C_{\text{ridge}}(\theta) \) using GD we only have adjust the gradient as follows
+$$
+\nabla_\theta C_{\text{ridge}}(\theta) = 2\begin{bmatrix} \sum_{i=1}^{100} \left(\theta_0+\theta_1x_i-y_i\right) \\
+\sum_{i=1}^{100}\left( x_i (\theta_0+\theta_1x_i)-y_ix_i\right) \\
+\end{bmatrix} + 2\lambda\begin{bmatrix} \theta_0 \\ \theta_1\end{bmatrix} = 2 (X^T(X\theta - \mathbf{y})+\lambda \theta).
+$$
+
+
+We can now extend our program to minimize \( C_{\text{ridge}}(\theta) \) using gradient descent and compare with the analytical solution given by
+$$
+\theta_{\text{ridge}} = \left(X^T X + \lambda I_{2 \times 2} \right)^{-1} X^T \mathbf{y},
+$$
+
+for \( \lambda = {0,1,10,50,100} \) (\( \lambda = 0 \) corresponds to ordinary least squares).
+We can then compute \( ||\theta_{\text{ridge}}|| \) for each \( \lambda \).
+
+
+
+
+
+
+Stochastic gradient descent (SGD) and variants thereof address some of
+the shortcomings of the Gradient descent method discussed above.
+
+
+The underlying idea of SGD comes from the observation that the cost
+function, which we want to minimize, can almost always be written as a
+sum over \( n \) datapoints \( \{\mathbf{x}_i\}_{i=1}^n \),
+$$
+C(\mathbf{\theta}) = \sum_{i=1}^n c_i(\mathbf{x}_i,
+\mathbf{\theta}).
+$$
+
+
+
+This in turn means that the gradient can be
+computed as a sum over \( i \)-gradients
+$$
+\nabla_\theta C(\mathbf{\theta}) = \sum_i^n \nabla_\theta c_i(\mathbf{x}_i,
+\mathbf{\theta}).
+$$
+
+
+Stochasticity/randomness is introduced by only taking the
+gradient on a subset of the data called minibatches. If there are \( n \)
+datapoints and the size of each minibatch is \( M \), there will be \( n/M \)
+minibatches. We denote these minibatches by \( B_k \) where
+\( k=1,\cdots,n/M \).
+
+
+
+The idea is now to approximate the gradient by replacing the sum over
+all datapoints with a sum over the datapoints in one the minibatches
+picked at random in each gradient descent step
+$$
+\nabla_\theta
+C(\mathbf{\theta}) = \sum_{i=1}^n \nabla_\theta c_i(\mathbf{x}_i,
+\mathbf{\theta}) \rightarrow \sum_{i \in B_k}^n \nabla_\theta
+c_i(\mathbf{x}_i, \mathbf{\theta}).
+$$
+
+
+
+Thus a gradient descent step now looks like
+$$
+\theta_{j+1} = \theta_j - \gamma_j \sum_{i \in B_k}^n \nabla_\theta c_i(\mathbf{x}_i,
+\mathbf{\theta})
+$$
+
+
+where \( k \) is picked at random with equal
+probability from \( [1,n/M] \). An iteration over the number of
+minibathces (n/M) is commonly referred to as an epoch. Thus it is
+typical to choose a number of epochs and for each epoch iterate over
+the number of minibatches, as exemplified in the code below.
+
+
+
+
+
+
+Taking the gradient only on a subset of the data has two important
+benefits. First, it introduces randomness which decreases the chance
+that our opmization scheme gets stuck in a local minima. Second, if
+the size of the minibatches are small relative to the number of
+datapoints (\( M < n \)), the computation of the gradient is much
+cheaper since we sum over the datapoints in the k-th minibatch and not
+all \( n \) datapoints.
+
+
+
+A natural question is when do we stop the search for a new minimum?
+One possibility is to compute the full gradient after a given number
+of epochs and check if the norm of the gradient is smaller than some
+threshold and stop if true. However, the condition that the gradient
+is zero is valid also for local minima, so this would only tell us
+that we are close to a local/global minimum. However, we could also
+evaluate the cost function at this point, store the result and
+continue the search. If the test kicks in at a later stage we can
+compare the values of the cost function and keep the \( \theta \) that
+gave the lowest value.
+
+
+
+Another approach is to let the step length \( \gamma_j \) depend on the
+number of epochs in such a way that it becomes very small after a
+reasonable time such that we do not move at all.
+
+
+As an example, let \( e = 0,1,2,3,\cdots \) denote the current epoch and let \( t_0, t_1 > 0 \) be two fixed numbers. Furthermore, let \( t = e \cdot m + i \) where \( m \) is the number of minibatches and \( i=0,\cdots,m-1 \). Then the function $$\gamma_j(t; t_0, t_1) = \frac{t_0}{t+t_1} $$ goes to zero as the number of epochs gets large. I.e. we start with a step length \( \gamma_j (0; t_0, t_1) = t_0/t_1 \) which decays in time \( t \).
+
+
+In this way we can fix the number of epochs, compute \( \theta \) and
+evaluate the cost function at the end. Repeating the computation will
+give a different result since the scheme is random by design. Then we
+pick the final \( \theta \) that gives the lowest value of the cost
+function.
+
+
+
+
+
+
+The success of the CG method for finding solutions of non-linear problems is based
+on the theory of conjugate gradients for linear systems of equations. It belongs
+to the class of iterative methods for solving problems from linear algebra of the type
+$$
+\begin{equation*}
+ \hat{A}\hat{x} = \hat{b}.
+\end{equation*}
+$$
+
+In the iterative process we end up with a problem like
+
+$$
+\begin{equation*}
+ \hat{r}= \hat{b}-\hat{A}\hat{x},
+\end{equation*}
+$$
+
+where \( \hat{r} \) is the so-called residual or error in the iterative process.
+
+
+When we have found the exact solution, \( \hat{r}=0 \).
+
+
+
+
+The residual is zero when we reach the minimum of the quadratic equation
+$$
+\begin{equation*}
+ P(\hat{x})=\frac{1}{2}\hat{x}^T\hat{A}\hat{x} - \hat{x}^T\hat{b},
+\end{equation*}
+$$
+
+with the constraint that the matrix \( \hat{A} \) is positive definite and symmetric.
+If we search for a minimum of the quantum mechanical variance, then the matrix
+\( \hat{A} \), which is called the Hessian, is given by the second-derivative of the function we want to minimize. This quantity is always positive definite. In our case this corresponds normally to the second derivative of the energy.
+
+
+We seek the minimum of the energy or the variance as function of various variational parameters.
+In our case we have thus a function \( f \) whose minimum we are seeking.
+In Newton's method we set \( \nabla f = 0 \) and we can thus compute the next iteration point
+$$
+\begin{equation*}
+\hat{x}-\hat{x}_i=\hat{A}^{-1}\nabla f(\hat{x}_i).
+\end{equation*}
+$$
+
+Subtracting this equation from that of \( \hat{x}_{i+1} \) we have
+$$
+\begin{equation*}
+\hat{x}_{i+1}-\hat{x}_i=\hat{A}^{-1}(\nabla f(\hat{x}_{i+1})-\nabla f(\hat{x}_i)).
+\end{equation*}
+$$
+
+
+The function \( f \) can be either the energy or the variance. If we choose the energy then we have
+$$
+\begin{equation*}
+\hat{\alpha}_{i+1}-\hat{\alpha}_i=\hat{A}^{-1}(\nabla E(\hat{\alpha}_{i+1})-\nabla E(\hat{\alpha}_i)).
+\end{equation*}
+$$
+
+In the simple harmonic oscillator model, the gradient and the Hessian \( \hat{A} \) are
+$$
+\begin{equation*}
+\frac{d\langle E_L[\alpha]\rangle}{d\alpha} = \alpha-\frac{1}{4\alpha^3}
+\end{equation*}
+$$
+
+and a second derivative which is always positive (meaning that we find a minimum)
+$$
+\begin{equation*}
+\hat{A}= \frac{d^2\langle E_L[\alpha]\rangle}{d\alpha^2} = 1+\frac{3}{4\alpha^4}
+\end{equation*}
+$$
+
+
+We get then
+$$
+\begin{equation*}
+\alpha_{i+1}=\frac{4}{3}\alpha_i-\frac{\alpha_i^4}{3\alpha_{i+1}^3},
+\end{equation*}
+$$
+
+which can be rewritten as
+$$
+\begin{equation*}
+\alpha_{i+1}^4-\frac{4}{3}\alpha_i\alpha_{i+1}^4+\frac{1}{3}\alpha_i^4.
+\end{equation*}
+$$
+
+
+In the CG method we define so-called conjugate directions and two vectors
+\( \hat{s} \) and \( \hat{t} \)
+are said to be
+conjugate if
+$$
+\begin{equation*}
+\hat{s}^T\hat{A}\hat{t}= 0.
+\end{equation*}
+$$
+
+The philosophy of the CG method is to perform searches in various conjugate directions
+of our vectors \( \hat{x}_i \) obeying the above criterion, namely
+$$
+\begin{equation*}
+\hat{x}_i^T\hat{A}\hat{x}_j= 0.
+\end{equation*}
+$$
+
+Two vectors are conjugate if they are orthogonal with respect to
+this inner product. Being conjugate is a symmetric relation: if \( \hat{s} \) is conjugate to \( \hat{t} \), then \( \hat{t} \) is conjugate to \( \hat{s} \).
+
+
+An example is given by the eigenvectors of the matrix
+$$
+\begin{equation*}
+\hat{v}_i^T\hat{A}\hat{v}_j= \lambda\hat{v}_i^T\hat{v}_j,
+\end{equation*}
+$$
+
+which is zero unless \( i=j \).
+
+
+Assume now that we have a symmetric positive-definite matrix \( \hat{A} \) of size
+\( n\times n \). At each iteration \( i+1 \) we obtain the conjugate direction of a vector
+$$
+\begin{equation*}
+\hat{x}_{i+1}=\hat{x}_{i}+\alpha_i\hat{p}_{i}.
+\end{equation*}
+$$
+
+We assume that \( \hat{p}_{i} \) is a sequence of \( n \) mutually conjugate directions.
+Then the \( \hat{p}_{i} \) form a basis of \( R^n \) and we can expand the solution
+$ \hat{A}\hat{x} = \hat{b}$ in this basis, namely
+
+$$
+\begin{equation*}
+ \hat{x} = \sum^{n}_{i=1} \alpha_i \hat{p}_i.
+\end{equation*}
+$$
+
+
+The coefficients are given by
+$$
+\begin{equation*}
+ \mathbf{A}\mathbf{x} = \sum^{n}_{i=1} \alpha_i \mathbf{A} \mathbf{p}_i = \mathbf{b}.
+\end{equation*}
+$$
+
+Multiplying with \( \hat{p}_k^T \) from the left gives
+
+$$
+\begin{equation*}
+ \hat{p}_k^T \hat{A}\hat{x} = \sum^{n}_{i=1} \alpha_i\hat{p}_k^T \hat{A}\hat{p}_i= \hat{p}_k^T \hat{b},
+\end{equation*}
+$$
+
+and we can define the coefficients \( \alpha_k \) as
+
+$$
+\begin{equation*}
+ \alpha_k = \frac{\hat{p}_k^T \hat{b}}{\hat{p}_k^T \hat{A} \hat{p}_k}
+\end{equation*}
+$$
+
+
+
+
+If we choose the conjugate vectors \( \hat{p}_k \) carefully,
+then we may not need all of them to obtain a good approximation to the solution
+\( \hat{x} \).
+We want to regard the conjugate gradient method as an iterative method.
+This will us to solve systems where \( n \) is so large that the direct
+method would take too much time.
+
+
+We denote the initial guess for \( \hat{x} \) as \( \hat{x}_0 \).
+We can assume without loss of generality that
+$$
+\begin{equation*}
+\hat{x}_0=0,
+\end{equation*}
+$$
+
+or consider the system
+$$
+\begin{equation*}
+\hat{A}\hat{z} = \hat{b}-\hat{A}\hat{x}_0,
+\end{equation*}
+$$
+
+instead.
+
+
+One can show that the solution \( \hat{x} \) is also the unique minimizer of the quadratic form
+$$
+\begin{equation*}
+ f(\hat{x}) = \frac{1}{2}\hat{x}^T\hat{A}\hat{x} - \hat{x}^T \hat{x} , \quad \hat{x}\in\mathbf{R}^n.
+\end{equation*}
+$$
+
+This suggests taking the first basis vector \( \hat{p}_1 \)
+to be the gradient of \( f \) at \( \hat{x}=\hat{x}_0 \),
+which equals
+$$
+\begin{equation*}
+\hat{A}\hat{x}_0-\hat{b},
+\end{equation*}
+$$
+
+and
+\( \hat{x}_0=0 \) it is equal \( -\hat{b} \).
+The other vectors in the basis will be conjugate to the gradient,
+hence the name conjugate gradient method.
+
+
+Let \( \hat{r}_k \) be the residual at the \( k \)-th step:
+$$
+\begin{equation*}
+\hat{r}_k=\hat{b}-\hat{A}\hat{x}_k.
+\end{equation*}
+$$
+
+Note that \( \hat{r}_k \) is the negative gradient of \( f \) at
+\( \hat{x}=\hat{x}_k \),
+so the gradient descent method would be to move in the direction \( \hat{r}_k \).
+Here, we insist that the directions \( \hat{p}_k \) are conjugate to each other,
+so we take the direction closest to the gradient \( \hat{r}_k \)
+under the conjugacy constraint.
+This gives the following expression
+$$
+\begin{equation*}
+\hat{p}_{k+1}=\hat{r}_k-\frac{\hat{p}_k^T \hat{A}\hat{r}_k}{\hat{p}_k^T\hat{A}\hat{p}_k} \hat{p}_k.
+\end{equation*}
+$$
+
+
+We can also compute the residual iteratively as
+$$
+\begin{equation*}
+\hat{r}_{k+1}=\hat{b}-\hat{A}\hat{x}_{k+1},
+ \end{equation*}
+$$
+
+which equals
+$$
+\begin{equation*}
+\hat{b}-\hat{A}(\hat{x}_k+\alpha_k\hat{p}_k),
+ \end{equation*}
+$$
+
+or
+$$
+\begin{equation*}
+(\hat{b}-\hat{A}\hat{x}_k)-\alpha_k\hat{A}\hat{p}_k,
+ \end{equation*}
+$$
+
+which gives
+
+$$
+\begin{equation*}
+\hat{r}_{k+1}=\hat{r}_k-\hat{A}\hat{p}_{k},
+ \end{equation*}
+$$
+
diff --git a/doc/pub/Splines/html/Splines.html b/doc/pub/Splines/html/Splines.html
index eec19b832..33591dd85 100644
--- a/doc/pub/Splines/html/Splines.html
+++ b/doc/pub/Splines/html/Splines.html
@@ -6,9 +6,9 @@ Automatically generated HTML file from DocOnce source
Cubic Splines
-Optimization, the central part of any Machine Learning algortithm
-
-Splines
-Steepest descent
+
$$
- s(x)=\begin{bmatrix} s_0(x)=a_0x+b_0 & x\in [x_0, x_1) \\
- s_1(x)=a_1x+b_1 & x\in [x_1, x_2) \\
- \dots & \dots \\
- s_{n-1}(x)=a_{n-1}x+b_{n-1} & x\in
- [x_{n-1}, x_n] \end{bmatrix}.
+\mathbf{x}_{k+1} = \mathbf{x}_k - \gamma_k \nabla F(\mathbf{x}_k), \ \ \gamma_k > 0
$$
-In this case the polynomial consists of series of straight lines
-connected to each other at every endpoint. The number of continuous
-derivatives is then \( k-1=0 \), as expected when we deal with straight lines.
-Such a polynomial is quite easy to construct given
-\( n+1 \) points \( x_0, x_1, \dots x_n \) and their corresponding
-function values.
-Splines
-More on Steepest descent
+
$$
- s_{i-1}(x_i)= y_i = s_i(x_i),
+\mathbf{x}_{k+1} = \mathbf{x}_k - \gamma_k \nabla F(\mathbf{x}_k), \ \ k \geq 0.
$$
-with \( 1 \le i \le n-1 \). In total we have \( n \) polynomials of the
-type
-
-$$
- s_i(x)=a_{i0}+a_{i1}x+a_{i2}x^2+a_{i2}x^3,
-$$
-
-
-yielding \( 4n \) coefficients to determine.
-Splines
-The ideal
+
-$$
- y_i = s(x_i),
-$$
-
+Ideally the sequence \( \{ \mathbf{x}_k \}_{k=0} \) converges to a global minimum of the function \( F \). In general we do not know if we are in a global or local minimum. In the special case when \( F \) is a convex function, all local minima are also global minima, so in this case gradient descent can converge to the global solution. The advantage of this scheme is that it is conceptually simple and straightforward to implement. However the method in this form has some severe limitations:
-and
-
-$$
- s(x_{i+1})= y_{i+1},
-$$
-
+
-$$
- s'_{i-1}(x_i)= s'_i(x_i),
-$$
-
-
-yields \( n-1 \) conditions. Similarly,
-
-$$
- s''_{i-1}(x_i)= s''_i(x_i),
-$$
-
-
-results in additional \( n-1 \) conditions. In total we have \( 4n \) coefficients
-and \( 4n-2 \) equations to determine them, leaving us with \( 2 \) degrees of
-freedom to be determined.
-Splines
-The sensitiveness of the gradient descent
+
-$$
- s''_{i}(x_i)= f_i,
-$$
-
+GD is sensitive to the choice of learning rate \( \gamma_k \). This is due
+to the fact that we are only guaranteed that \( F(\mathbf{x}_{k+1}) \leq
+F(\mathbf{x}_k) \) for sufficiently small \( \gamma_k \). The problem is to
+determine an optimal learning rate. If the learning rate is chosen to
+small the method will take a long to converge and if it is to large we
+can experience erratic behavior.
-and
-
-$$
- s''_{i}(x_{i+1})= f_{i+1},
-$$
-
-
-and setting up a straight line between \( f_i \) and \( f_{i+1} \) we have
-
-$$
- s_i''(x) = \frac{f_i}{x_{i+1}-x_i}(x_{i+1}-x)+
- \frac{f_{i+1}}{x_{i+1}-x_i}(x-x_i),
-$$
-
-
-and integrating twice one obtains
-
-$$
- s_i(x) = \frac{f_i}{6(x_{i+1}-x_i)}(x_{i+1}-x)^3+
- \frac{f_{i+1}}{6(x_{i+1}-x_i)}(x-x_i)^3
- +c(x-x_i)+d(x_{i+1}-x).
-$$
-
-Splines
-Gradient Descent Example
+We revisit now our simple linear regression example with a linear polynomial.
# Importing various packages
+from random import random, seed
+import numpy as np
+import matplotlib.pyplot as plt
+from mpl_toolkits.mplot3d import Axes3D
+from matplotlib import cm
+from matplotlib.ticker import LinearLocator, FormatStrFormatter
+import sys
+
+x = 2*np.random.rand(100,1)
+y = 4+3*x+np.random.randn(100,1)
+
+xb = np.c_[np.ones((100,1)), x]
+theta_linreg = np.linalg.inv(xb.T.dot(xb)).dot(xb.T).dot(y)
+print(theta_linreg)
+theta = np.random.randn(2,1)
+
+eta = 0.1
+Niterations = 1000
+m = 100
+
+for iter in range(Niterations):
+ gradients = 2.0/m*xb.T.dot(xb.dot(theta)-y)
+ theta -= eta*gradients
+
+print(theta)
+xnew = np.array([[0],[2]])
+xbnew = np.c_[np.ones((2,1)), xnew]
+ypredict = xbnew.dot(theta)
+ypredict2 = xbnew.dot(theta_linreg)
+plt.plot(xnew, ypredict, "r-")
+plt.plot(xnew, ypredict2, "b-")
+plt.plot(x, y ,'ro')
+plt.axis([0,2.0,0, 15.0])
+plt.xlabel(r'$x$')
+plt.ylabel(r'$y$')
+plt.title(r'Random numbers ')
+plt.show()
+
And a corresponding example using scikit-learn
+
+# Importing various packages
+from random import random, seed
+import numpy as np
+import matplotlib.pyplot as plt
+from sklearn.linear_model import SGDRegressor
+
+x = 2*np.random.rand(100,1)
+y = 4+3*x+np.random.randn(100,1)
+
+xb = np.c_[np.ones((100,1)), x]
+theta_linreg = np.linalg.inv(xb.T.dot(xb)).dot(xb.T).dot(y)
+print(theta_linreg)
+sgdreg = SGDRegressor(n_iter = 50, penalty=None, eta0=0.1)
+sgdreg.fit(x,y.ravel())
+print(sgdreg.intercept_, sgdreg.coef_)
+
Convex functions
+Ideally we want our cost/loss function to be convex(concave).
+
+Convex function
+
+
+$$f(tx_1 + (1-t)x_2) \leq tf(x_1) + (1-t)f(x_2) $$
+
for all \( x_1, x_2 \in X \) and for all \( t \in [0,1] \). If \( \leq \) is replaced with a strict inequaltiy in the definition, we demand \( x_1 \neq x_2 \) and \( t\in(0,1) \) then \( f \) is said to be strictly convex. For a single variable function, convexity means that if you draw a straight line connecting \( f(x_1) \) and \( f(x_2) \), the value of the function on the interval \( [x_1,x_2] \) is always below the line as illustrated below.
+Conditions on convex functions
+
+
+$$f(y) \geq f(x) + \nabla f(x)^T (y-x) $$
+
holds
+for all \( x,y \in D_f \). This condition means that for a convex function
+the first order Taylor expansion (right hand side above) at any point
+a global under estimator of the function. To convince yourself you can
+make a drawing of f(x) = x^2+1 and draw the tangent line to \( f(x) \) and
+note that it is always below the graph.
+More on convex functions
+
+Some simple problems
+
+
+
+
+$$\lambda f(x) + (1-\lambda)f(y) - f(\lambda x + (1-\lambda) y ) \geq 0. $$
+
+
+
+
+
+
+Revisiting our first homework
+
+
+
+
$$
-\begin{align}
- s_i(x) =&\frac{f_i}{6(x_{i+1}-x_i)}(x_{i+1}-x)^3+
- \frac{f_{i+1}}{6(x_{i+1}-x_i)}(x-x_i)^3 \nonumber \\
- +&(\frac{y_{i+1}}{x_{i+1}-x_i}-\frac{f_{i+1}(x_{i+1}-x_i)}{6})
- (x-x_i)+
- (\frac{y_{i}}{x_{i+1}-x_i}-\frac{f_{i}(x_{i+1}-x_i)}{6})
- (x_{i+1}-x).
+y_i = 5x_i^2 + 0.1\xi_i, \ i=1,\cdots,100
+$$
+
+
+with \( x_i \in [0,1] \) chosen randomly with a uniform distribution. Additionally \( \xi_i \) represents stochastic noise chosen according to a normal distribution \( \cal {N}(0,1) \).
+The linear regression model is given by
+
+$$
+h_\theta(x) = \hat{y} = \theta_0 + \theta_1 x,
+$$
+
+
+such that
+
+$$
+\hat{y}_i = \theta_0 + \theta_1 x_i.
+$$
+
+Gradient descent example
+
+
+$$
+\begin{equation}
+X \equiv \begin{bmatrix}
+1 & x_1 \\
+\vdots & \vdots \\
+1 & x_{100} & \\
+\end{bmatrix}.
\tag{1}
-\end{align}
+\end{equation}
$$
-
+$$
+C(\theta) = ||X\theta-\mathbf{y}||^2 = ||X\theta||^2 - 2 \mathbf{y}^T X\theta + ||\mathbf{y}||^2 = \sum_{i=1}^{100} (\theta_0 + \theta_1 x_i)^2 - 2 y_i (\theta_0 + \theta_1 x_i) + y_i^2
+$$
+
+
+and we want to find \( \theta \) such that \( C(\theta) \) is minimized.
Splines
-The derivative of the cost/loss function
+
$$
- s'_{i-1}(x_i)= s'_i(x_i),
+\nabla_\theta C(\theta) = (\partial C(\theta) / \partial \theta_0, \partial C(\theta) / \partial \theta_1)^T = 2\begin{bmatrix} \sum_{i=1}^{100} \left(\theta_0+\theta_1x_i-y_i\right) \\
+\sum_{i=1}^{100}\left( x_i (\theta_0+\theta_1x_i)-y_ix_i\right) \\
+\end{bmatrix} = 2X^T(X\theta - \mathbf{y}),
$$
-and set \( x=x_i \). Defining \( h_i=x_{i+1}-x_i \) we obtain finally
-the following expression
-
-$$
- h_{i-1}f_{i-1}+2(h_{i}+h_{i-1})f_i+h_if_{i+1}=
- \frac{6}{h_i}(y_{i+1}-y_i)-\frac{6}{h_{i-1}}(y_{i}-y_{i-1}),
-$$
-
-
-and introducing the shorthands \( u_i=2(h_{i}+h_{i-1}) \),
-\( v_i=\frac{6}{h_i}(y_{i+1}-y_i)-\frac{6}{h_{i-1}}(y_{i}-y_{i-1}) \),
-we can reformulate the problem as a set of linear equations to be
-solved through e.g., Gaussian elemination
-Splines
-The Hessian matrix
+The Hessian matrix of \( C(\theta) \) is given by
$$
- \begin{bmatrix} u_1 & h_1 &0 &\dots & & & & \\
- h_1 & u_2 & h_2 &0 &\dots & & & \\
- 0 & h_2 & u_3 & h_3 &0 &\dots & & \\
- \dots& & \dots &\dots &\dots &\dots &\dots & \\
- &\dots & & &0 &h_{n-3} &u_{n-2} &h_{n-2} \\
- & && & &0 &h_{n-2} &u_{n-1} \end{bmatrix}
- \begin{bmatrix} f_1 \\
- f_2 \\
- f_3\\
- \dots \\
- f_{n-2} \\
- f_{n-1} \end{bmatrix} =
- \begin{bmatrix} v_1 \\
- v_2 \\
- v_3\\
- \dots \\
- v_{n-2}\\
- v_{n-1} \end{bmatrix}.
+\hat{H} \equiv \begin{bmatrix}
+\frac{\partial^2 C(\theta)}{\partial \theta_0^2} & \frac{\partial^2 C(\theta)}{\partial \theta_0 \partial \theta_1} \\
+\frac{\partial^2 C(\theta)}{\partial \theta_0 \partial \theta_1} & \frac{\partial^2 C(\theta)}{\partial \theta_1^2} & \\
+\end{bmatrix} = 2X^T X.
$$
-Note that this is a set of tridiagonal equations and can be solved
-through only \( O(n) \) operations.
-Splines
-Simple program
+
+$$
+\theta_{k+1} = \theta_k - \gamma \nabla_\theta C(\theta_k), \ k=0,1,\cdots
+$$
+
+
+import numpy as np
+
+"""
+The following setup is just a suggestion, feel free to write it the way you like.
+"""
+
+#Setup problem described in the exercise
+N = 100 #Nr of datapoints
+M = 2 #Nr of features
+x = np.random.rand(N) #Uniformly generated x-values in [0,1]
+y = 5*x**2 + 0.1*np.random.randn(N)
+X = np.c_[np.ones(N),x] #Construct design matrix
+
+#Compute theta according to normal equations to compare with GD solution
+Xt_X_inv = np.linalg.inv(np.dot(X.T,X))
+Xt_y = np.dot(X.transpose(),y)
+theta_NE = np.dot(Xt_X_inv,Xt_y)
+print(theta_NE)
+
Gradient descent and Ridge
+
+
+$$
+C_{\text{ridge}}(\theta) = ||X\theta -\mathbf{y}||^2 + \lambda ||\theta||^2, \ \lambda \geq 0.
+$$
+
+
+
+$$
+\nabla_\theta C_{\text{ridge}}(\theta) = 2\begin{bmatrix} \sum_{i=1}^{100} \left(\theta_0+\theta_1x_i-y_i\right) \\
+\sum_{i=1}^{100}\left( x_i (\theta_0+\theta_1x_i)-y_ix_i\right) \\
+\end{bmatrix} + 2\lambda\begin{bmatrix} \theta_0 \\ \theta_1\end{bmatrix} = 2 (X^T(X\theta - \mathbf{y})+\lambda \theta).
+$$
+
+
+
+$$
+\theta_{\text{ridge}} = \left(X^T X + \lambda I_{2 \times 2} \right)^{-1} X^T \mathbf{y},
+$$
+
+
+for \( \lambda = {0,1,10,50,100} \) (\( \lambda = 0 \) corresponds to ordinary least squares).
+We can then compute \( ||\theta_{\text{ridge}}|| \) for each \( \lambda \).
spline(double x[], double y[], int n, double yp1, double yp2, double y2[])
+
+
import numpy as np
+
+"""
+The following setup is just a suggestion, feel free to write it the way you like.
+"""
+
+#Setup problem described in the exercise
+N = 100 #Nr of datapoints
+M = 2 #Nr of features
+x = np.random.rand(N)
+y = 5*x**2 + 0.1*np.random.randn(N)
+
+
+#Compute analytic theta for Ridge regression
+X = np.c_[np.ones(N),x]
+XT_X = np.dot(X.T,X)
+
+l = 0.1 #Ridge parameter lambda
+Id = np.eye(XT_X.shape[0])
+
+Z = np.linalg.inv(XT_X+l*Id)
+theta_ridge = np.dot(Z,np.dot(X.T,y))
+
+print(theta_ridge)
+print(np.linalg.norm(theta_ridge)) #||theta||
+
Stochastic Gradient Descent
+
+
+$$
+C(\mathbf{\theta}) = \sum_{i=1}^n c_i(\mathbf{x}_i,
+\mathbf{\theta}).
+$$
+
+Computation of gradients
+
+
+$$
+\nabla_\theta C(\mathbf{\theta}) = \sum_i^n \nabla_\theta c_i(\mathbf{x}_i,
+\mathbf{\theta}).
+$$
+
+
+SGD example
+As an example, suppose we have \( 10 \) datapoints \( ( \mathbf{x}_1,
+\cdots, \mathbf{x}_{10} ) \) and we choose to have \( M=5 \) minibathces,
+then each minibatch contains two datapoints. In particular we have
+\( B_1 = (\mathbf{x}_1,\mathbf{x}_2), \cdots, B_5 =
+(\mathbf{x}_9,\mathbf{x}_{10}) \). Note that if you choose \( M=1 \) you
+have only a single batch with all datapoints and on the other extreme,
+you may choose \( M=n \) resulting in a minibatch for each datapoint, i.e
+\( B_k = \mathbf{x}_k \).
+
+
+$$
+\nabla_\theta
+C(\mathbf{\theta}) = \sum_{i=1}^n \nabla_\theta c_i(\mathbf{x}_i,
+\mathbf{\theta}) \rightarrow \sum_{i \in B_k}^n \nabla_\theta
+c_i(\mathbf{x}_i, \mathbf{\theta}).
+$$
+
+The gradient step
+
+
+$$
+\theta_{j+1} = \theta_j - \gamma_j \sum_{i \in B_k}^n \nabla_\theta c_i(\mathbf{x}_i,
+\mathbf{\theta})
+$$
+
+
+Simple example code
+
+import numpy as np
+
+n = 100 #100 datapoints
+M = 5 #size of each minibatch
+m = int(n/M) #number of minibatches
+n_epochs = 10 #number of epochs
+
+j = 0
+for epoch in range(1,n_epochs+1):
+ for i in range(m):
+ k = np.random.randint(m) #Pick the k-th minibatch at random
+ #Compute the gradient using the data in minibatch Bk
+ #Compute new suggestion for theta
+ j += 1
Splines
-When do we stop?
+
Slightly different approach
+
+
+$$\gamma_j(t; t_0, t_1) = \frac{t_0}{t+t_1} $$
+
goes to zero as the number of epochs gets large. I.e. we start with a step length \( \gamma_j (0; t_0, t_1) = t_0/t_1 \) which decays in time \( t \).
+
+splint(double x[], double y[], double y2a[], int n, double x, double *y)
+
+
import numpy as np
+
+def step_length(t,t0,t1):
+ return t0/(t+t1)
+
+n = 100 #100 datapoints
+M = 5 #size of each minibatch
+m = int(n/M) #number of minibatches
+n_epochs = 500 #number of epochs
+t0 = 1.0
+t1 = 10
+
+gamma_j = t0/t1
+j = 0
+for epoch in range(1,n_epochs+1):
+ for i in range(m):
+ k = np.random.randint(m) #Pick the k-th minibatch at random
+ #Compute the gradient using the data in minibatch Bk
+ #Compute new suggestion for theta
+ t = epoch*m+i
+ gamma_j = step_length(t,t0,t1)
+ j += 1
+
+print("gamma_j after %d epochs: %g" % (n_epochs,gamma_j))
Conjugate gradient (CG) method
+Conjugate gradient (CG) method
Conjugate gradient method
+Conjugate gradient method
Conjugate gradient method, Newton's method first
+Conjugate gradient method, Newton's method first
Simple example and demonstration
+Simple example and demonstration
Simple example and demonstration
+Simple example and demonstration
Conjugate gradient method
+Conjugate gradient method
Conjugate gradient method
+Conjugate gradient method
Conjugate gradient method
+Conjugate gradient method
Conjugate gradient method
+Conjugate gradient method
Conjugate gradient method and iterations
+Conjugate gradient method and iterations
Conjugate gradient method
+Conjugate gradient method
Conjugate gradient method
+Conjugate gradient method
Conjugate gradient method
+Conjugate gradient method
Gradient Descent codes
-# Importing various packages
-from math import exp, sqrt
-from random import random, seed
-import numpy as np
-import matplotlib.pyplot as plt
-from mpl_toolkits.mplot3d import Axes3D
-from matplotlib import cm
-from matplotlib.ticker import LinearLocator, FormatStrFormatter
-import sys
-
-x = 2*np.random.rand(100,1)
-y = 4+3*x+np.random.randn(100,1)
-
-xb = np.c_[np.ones((100,1)), x]
-theta_linreg = np.linalg.inv(xb.T.dot(xb)).dot(xb.T).dot(y)
-print(theta_linreg)
-theta = np.random.randn(2,1)
-
-eta = 0.1
-Niterations = 1000
-m = 100
-
-for iter in range(Niterations):
- gradients = 2.0/m*xb.T.dot(xb.dot(theta)-y)
- theta -= eta*gradients
-
-print(theta)
-xnew = np.array([[0],[2]])
-xbnew = np.c_[np.ones((2,1)), xnew]
-ypredict = xbnew.dot(theta)
-ypredict2 = xbnew.dot(theta_linreg)
-plt.plot(xnew, ypredict, "r-")
-plt.plot(xnew, ypredict2, "b-")
-plt.plot(x, y ,'ro')
-plt.axis([0,2.0,0, 15.0])
-plt.xlabel(r'$x$')
-plt.ylabel(r'$y$')
-plt.title(r'Random numbers ')
-plt.show()
-
# Importing various packages
-from math import exp, sqrt
-from random import random, seed
-import numpy as np
-import matplotlib.pyplot as plt
-from sklearn.linear_model import SGDRegressor
-
-x = 2*np.random.rand(100,1)
-y = 4+3*x+np.random.randn(100,1)
-
-xb = np.c_[np.ones((100,1)), x]
-theta_linreg = np.linalg.inv(xb.T.dot(xb)).dot(xb.T).dot(y)
-print(theta_linreg)
-sgdreg = SGDRegressor(n_iter = 50, penalty=None, eta0=0.1)
-sgdreg.fit(x,y.ravel())
-print(sgdreg.intercept_, sgdreg.coef_)
-
Data Analysis and Machine Learning Lectures: Cubic Splines and Gradient Methods
Data Analysis and Machine Learning Lectures: Optimization and Gradient Methods
May 22, 2018
Sep 20, 2018
-Cubic Splines
-Optimization, the central part of any Machine Learning algortithm
-
-
-Splines
-Steepest descent
+
More on Steepest descent
+
+The ideal
+
+The sensitiveness of the gradient descent
+
+
-Splines
-
-
-Splines
-
-
-Splines
-
-
-Splines
-
-
-Splines
-
-
-Splines
-
-
-Splines
-spline(double x[], double y[], int n, double yp1, double yp2, double y2[])
-
-
-Splines
-splint(double x[], double y[], double y2a[], int n, double x, double *y)
-
-
-Conjugate gradient (CG) method
-
-
-Conjugate gradient method
-
-
-Conjugate gradient method, Newton's method first
-
-
-Simple example and demonstration
-
-
-Simple example and demonstration
-
-
-Conjugate gradient method
-
-
-Conjugate gradient method
-
-
-Conjugate gradient method
-
-
-Conjugate gradient method
-
-
-Conjugate gradient method and iterations
-
-
-Conjugate gradient method
-
-
-Conjugate gradient method
-
-
-Conjugate gradient method
-
-
-Gradient Descent codes
+Gradient Descent Example
+We revisit now our simple linear regression example with a linear polynomial.
# Importing various packages
-from math import exp, sqrt
from random import random, seed
import numpy as np
import matplotlib.pyplot as plt
@@ -831,11 +287,15 @@ plt.ylabel(r'$y$')
plt.title(r'Random numbers ')
plt.show()
+
+And a corresponding example using scikit-learn
+
# Importing various packages
-from math import exp, sqrt
from random import random, seed
import numpy as np
import matplotlib.pyplot as plt
@@ -851,6 +311,864 @@ sgdreg = SGDRegressor(n_iter = 50, penalty=<
sgdreg.fit(x,y.ravel())
print(sgdreg.intercept_, sgdreg.coef_)
Convex functions
+Ideally we want our cost/loss function to be convex(concave).
+
+
+
+Convex function
+
+
+
+Conditions on convex functions
+
+
+
+More on convex functions
+
+
+
+Some simple problems
+
+
+
+
+Hint: If you re-write the definition, \( f \) is convex if the following holds for all \( x,y \in D_f \) and any \( \lambda \in [0,1] \) $$\lambda f(x) + (1-\lambda)f(y) - f(\lambda x + (1-\lambda) y ) \geq 0. $$
+
+
+
+
+\( f(x) = e^x \) is convex for \( x \in \mathbb{R} \).
+\( g(x) = -\ln(x) \) is convex for \( x \in (0,\infty) \).
+
+
+
+
+\( f(\alpha x) = |\alpha| f(x) \) for all \( \alpha \in \mathbb{R} \).
+\( f(x+y) \leq f(x) + f(y) \)
+\( f(x) \leq 0 \) for all \( x \in \mathbb{R}^n \) with equality if and only if \( x = 0 \)
+Using the definition of convexity, show that a function satisfying the properties above is convex (the third condition is not needed to show this).
+
+Revisiting our first homework
+
+
+
+
+We revisit the example from homework set 1 where we had
+$$
+y_i = 5x_i^2 + 0.1\xi_i, \ i=1,\cdots,100
+$$
+
+with \( x_i \in [0,1] \) chosen randomly with a uniform distribution. Additionally \( \xi_i \) represents stochastic noise chosen according to a normal distribution \( \cal {N}(0,1) \).
+The linear regression model is given by
+$$
+h_\theta(x) = \hat{y} = \theta_0 + \theta_1 x,
+$$
+
+such that
+$$
+\hat{y}_i = \theta_0 + \theta_1 x_i.
+$$
+
+Gradient descent example
+
+
+
+The derivative of the cost/loss function
+
+
+
+The Hessian matrix
+The Hessian matrix of \( C(\theta) \) is given by
+$$
+\hat{H} \equiv \begin{bmatrix}
+\frac{\partial^2 C(\theta)}{\partial \theta_0^2} & \frac{\partial^2 C(\theta)}{\partial \theta_0 \partial \theta_1} \\
+\frac{\partial^2 C(\theta)}{\partial \theta_0 \partial \theta_1} & \frac{\partial^2 C(\theta)}{\partial \theta_1^2} & \\
+\end{bmatrix} = 2X^T X.
+$$
+
+This result implies that \( C(\theta) \) is a convex function since the matrix \( X^T X \) always is positive semi-definite.
+
+
+
+Simple program
+
+import numpy as np
+
+"""
+The following setup is just a suggestion, feel free to write it the way you like.
+"""
+
+#Setup problem described in the exercise
+N = 100 #Nr of datapoints
+M = 2 #Nr of features
+x = np.random.rand(N) #Uniformly generated x-values in [0,1]
+y = 5*x**2 + 0.1*np.random.randn(N)
+X = np.c_[np.ones(N),x] #Construct design matrix
+
+#Compute theta according to normal equations to compare with GD solution
+Xt_X_inv = np.linalg.inv(np.dot(X.T,X))
+Xt_y = np.dot(X.transpose(),y)
+theta_NE = np.dot(Xt_X_inv,Xt_y)
+print(theta_NE)
+
Gradient descent and Ridge
+
+import numpy as np
+
+"""
+The following setup is just a suggestion, feel free to write it the way you like.
+"""
+
+#Setup problem described in the exercise
+N = 100 #Nr of datapoints
+M = 2 #Nr of features
+x = np.random.rand(N)
+y = 5*x**2 + 0.1*np.random.randn(N)
+
+
+#Compute analytic theta for Ridge regression
+X = np.c_[np.ones(N),x]
+XT_X = np.dot(X.T,X)
+
+l = 0.1 #Ridge parameter lambda
+Id = np.eye(XT_X.shape[0])
+
+Z = np.linalg.inv(XT_X+l*Id)
+theta_ridge = np.dot(Z,np.dot(X.T,y))
+
+print(theta_ridge)
+print(np.linalg.norm(theta_ridge)) #||theta||
+
+
+Stochastic Gradient Descent
+
+
+
+Computation of gradients
+
+
+
+SGD example
+As an example, suppose we have \( 10 \) datapoints \( ( \mathbf{x}_1,
+\cdots, \mathbf{x}_{10} ) \) and we choose to have \( M=5 \) minibathces,
+then each minibatch contains two datapoints. In particular we have
+\( B_1 = (\mathbf{x}_1,\mathbf{x}_2), \cdots, B_5 =
+(\mathbf{x}_9,\mathbf{x}_{10}) \). Note that if you choose \( M=1 \) you
+have only a single batch with all datapoints and on the other extreme,
+you may choose \( M=n \) resulting in a minibatch for each datapoint, i.e
+\( B_k = \mathbf{x}_k \).
+
+
+
+The gradient step
+
+
+
+Simple example code
+
+import numpy as np
+
+n = 100 #100 datapoints
+M = 5 #size of each minibatch
+m = int(n/M) #number of minibatches
+n_epochs = 10 #number of epochs
+
+j = 0
+for epoch in range(1,n_epochs+1):
+ for i in range(m):
+ k = np.random.randint(m) #Pick the k-th minibatch at random
+ #Compute the gradient using the data in minibatch Bk
+ #Compute new suggestion for theta
+ j += 1
+
+
+When do we stop?
+
+
+
+Slightly different approach
+
+import numpy as np
+
+def step_length(t,t0,t1):
+ return t0/(t+t1)
+
+n = 100 #100 datapoints
+M = 5 #size of each minibatch
+m = int(n/M) #number of minibatches
+n_epochs = 500 #number of epochs
+t0 = 1.0
+t1 = 10
+
+gamma_j = t0/t1
+j = 0
+for epoch in range(1,n_epochs+1):
+ for i in range(m):
+ k = np.random.randint(m) #Pick the k-th minibatch at random
+ #Compute the gradient using the data in minibatch Bk
+ #Compute new suggestion for theta
+ t = epoch*m+i
+ gamma_j = step_length(t,t0,t1)
+ j += 1
+
+print("gamma_j after %d epochs: %g" % (n_epochs,gamma_j))
+
+
+Conjugate gradient (CG) method
+
+
+Conjugate gradient method
+
+
+Conjugate gradient method, Newton's method first
+
+
+Simple example and demonstration
+
+
+Simple example and demonstration
+
+
+Conjugate gradient method
+
+
+Conjugate gradient method
+
+
+Conjugate gradient method
+
+
+Conjugate gradient method
+
+
+Conjugate gradient method and iterations
+
+
+Conjugate gradient method
+
+
+Conjugate gradient method
+
+
+Conjugate gradient method
+