Data Analysis and Machine Learning:

Logistic Regression and Gradient

Methods

Morten Hjorth-Jensen1,2
1
Department of Physics and Center for Computing in Science Education, University of Oslo, Norway
partment of Physics and Astronomy and Facility for Rare Ion Beams and National Superconducting Cyclotron Laboratory, Michigan State University, U

Jun 26, 2020

Logistic Regression
In linear regression our main interest was centered on learning the coefficients of
a functional fit (say a polynomial) in order to be able to predict the response of a
continuous variable on some unseen data. The fit to the continuous variable yi is
based on some independent variables xi . Linear regression resulted in analytical
expressions for standard ordinary Least Squares or Ridge regression (in terms of
matrices to invert) for several quantities, ranging from the variance and thereby
the confidence intervals of the parameters β to the mean squared error. If we can
invert the product of the design matrices, linear regression gives then a simple
recipe for fitting our data.

Logistic Regression and Classification Problems

Classification problems, however, are concerned with outcomes taking the form
of discrete variables (i.e. categories). We may for example, on the basis of
DNA sequencing for a number of patients, like to find out which mutations are
important for a certain disease; or based on scans of various patients’ brains,
figure out if there is a tumor or not; or given a specific physical system, we’d like
to identify its state, say whether it is an ordered or disordered system (typical
situation in solid state physics); or classify the status of a patient, whether
she/he has a stroke or not and many other similar situations.
The most common situation we encounter when we apply logistic regression
is that of two possible outcomes, normally denoted as a binary outcome, true or
false, positive or negative, success or failure etc.

c 1999-2020, Morten Hjorth-Jensen. Released under CC Attribution-NonCommercial 4.0

license
Optimization and Deep learning
Logistic regression will also serve as our stepping stone towards neural network
algorithms and supervised deep learning. For logistic learning, the minimization
of the cost function leads to a non-linear equation in the parameters β. The
optimization of the problem calls therefore for minimization algorithms. This
forms the bottle neck of all machine learning algorithms, namely how to find
reliable minima of a multi-variable function. This leads us to the family of
gradient descent methods. The latter are the working horses of basically all
modern machine learning algorithms.
We note also that many of the topics discussed here on logistic regression
are also commonly used in modern supervised Deep Learning models, as we will
see later.

Basics
We consider the case where the dependent variables, also called the responses or
the outcomes, yi are discrete and only take values from k = 0, . . . , K − 1 (i.e. K
classes).
The goal is to predict the output classes from the design matrix X ∈ Rn×p
made of n samples, each of which carries p features or predictors. The primary
goal is to identify the classes to which new unseen samples belong.
Let us specialize to the case of two classes only, with outputs yi = 0 and
yi = 1. Our outcomes could represent the status of a credit card user that could
default or not on her/his credit card debt. That is
 
0 no
yi = .
1 yes

Linear classifier
Before moving to the logistic model, let us try to use our linear regression model
to classify these two outcomes. We could for example fit a linear model to the
default case if yi > 0.5 and the no default case yi ≤ 0.5.
We would then have our weighted linear combination, namely
y = Xβ + , (1)
where y is a vector representing the possible outcomes, X is our n × p design
matrix and β represents our estimators/predictors.

Some selected properties

The main problem with our function is that it takes values on the entire real axis.
In the case of logistic regression, however, the labels yi are discrete variables. A
typical example is the credit card data discussed earlier, where we can set the
state of defaulting the debt to yi = 1 and not to yi = 0 for one the persons in
the data set (see the full example below).

2
 One simple way to get a discrete output is to have sign functions that map
the output of a linear regressor to values {0, 1}, f (si ) = sign(si ) = 1 if si ≥ 0
and 0 if otherwise. We will encounter this model in our first demonstration of
neural networks. Historically it is called the “perceptron" model in the machine
learning literature. This model is extremely simple. However, in many cases it
is more favorable to use a “soft" classifier that outputs the probability of a given
category. This leads us to the logistic function.

The logistic function

The perceptron is an example of a “hard classification” model. We will encounter
this model when we discuss neural networks as well. Each datapoint is deter-
ministically assigned to a category (i.e yi = 0 or yi = 1). In many cases, it
is favorable to have a “soft” classifier that outputs the probability of a given
category rather than a single value. For example, given xi , the classifier outputs
the probability of being in a category k. Logistic regression is the most common
example of a so-called soft classifier. In logistic regression, the probability that
a data point xi belongs to a category yi = {0, 1} is given by the so-called logit
function (or Sigmoid) which is meant to represent the likelihood for a given
event,
1 exp t
p(t) = = .
1 + exp−t 1 + expt
Note that 1 − p(t) = p(−t).

Examples of likelihood functions used in logistic regression

and nueral networks
The following code plots the logistic function, the step function and other
functions we will encounter from here and on.
"""The sigmoid function (or the logistic curve) is a
function that takes any real number, z, and outputs a number (0,1).
It is useful in neural networks for assigning weights on a relative scale.
The value z is the weighted sum of parameters involved in the learning algorithm."""

import numpy
import matplotlib.pyplot as plt
import math as mt
z = numpy.arange(-5, 5, .1)
sigma_fn = numpy.vectorize(lambda z: 1/(1+numpy.exp(-z)))
sigma = sigma_fn(z)

fig = plt.figure()
ax = fig.add_subplot(111)
ax.plot(z, sigma)
ax.set_ylim([-0.1, 1.1])
ax.set_xlim([-5,5])
ax.grid(True)
ax.set_xlabel(’z’)
ax.set_title(’sigmoid function’)

3
 plt.show()

"""Step Function"""
z = numpy.arange(-5, 5, .02)
step_fn = numpy.vectorize(lambda z: 1.0 if z >= 0.0 else 0.0)
step = step_fn(z)

fig = plt.figure()
ax = fig.add_subplot(111)
ax.plot(z, step)
ax.set_ylim([-0.5, 1.5])
ax.set_xlim([-5,5])
ax.grid(True)
ax.set_xlabel(’z’)
ax.set_title(’step function’)

plt.show()

"""tanh Function"""
z = numpy.arange(-2*mt.pi, 2*mt.pi, 0.1)
t = numpy.tanh(z)

fig = plt.figure()
ax = fig.add_subplot(111)
ax.plot(z, t)
ax.set_ylim([-1.0, 1.0])
ax.set_xlim([-2*mt.pi,2*mt.pi])
ax.grid(True)
ax.set_xlabel(’z’)
ax.set_title(’tanh function’)

plt.show()

Two parameters
We assume now that we have two classes with yi either 0 or 1. Furthermore we
assume also that we have only two parameters β in our fitting of the Sigmoid
function, that is we define probabilities

exp (β0 + β1 xi )
p(yi = 1|xi , β) = ,
1 + exp (β0 + β1 xi )
p(yi = 0|xi , β) = 1 − p(yi = 1|xi , β),

where β are the weights we wish to extract from data, in our case β0 and β1 .
Note that we used

p(yi = 0|xi , β) = 1 − p(yi = 1|xi , β).

Maximum likelihood
In order to define the total likelihood for all possible outcomes from a dataset
D = {(yi , xi )}, with the binary labels yi ∈ {0, 1} and where the data points
are drawn independently, we use the so-called Maximum Likelihood Estimation

4
(MLE) principle. We aim thus at maximizing the probability of seeing the
observed data. We can then approximate the likelihood in terms of the product
of the individual probabilities of a specific outcome yi , that is
n
Y y 1−yi
P (D|β) = [p(yi = 1|xi , β)] i [1 − p(yi = 1|xi , β))]
i=1

from which we obtain the log-likelihood and our cost/loss function

n
X
C(β) = (yi log p(yi = 1|xi , β) + (1 − yi ) log [1 − p(yi = 1|xi , β))]) .
i=1

The cost function rewritten

Reordering the logarithms, we can rewrite the cost/loss function as
n
X
C(β) = (yi (β0 + β1 xi ) − log (1 + exp (β0 + β1 xi ))) .
i=1

The maximum likelihood estimator is defined as the set of parameters that

maximize the log-likelihood where we maximize with respect to β. Since the
cost (error) function is just the negative log-likelihood, for logistic regression we
have that
n
X
C(β) = − (yi (β0 + β1 xi ) − log (1 + exp (β0 + β1 xi ))) .
i=1

This equation is known in statistics as the cross entropy. Finally, we note that
just as in linear regression, in practice we often supplement the cross-entropy
with additional regularization terms, usually L1 and L2 regularization as we did
for Ridge and Lasso regression.

Minimizing the cross entropy

The cross entropy is a convex function of the weights β and, therefore, any local
minimizer is a global minimizer.
Minimizing this cost function with respect to the two parameters β0 and β1
we obtain
n  
∂C(β) X exp (β0 + β1 xi )
=− yi − ,
∂β0 i=1
1 + exp (β0 + β1 xi )
and
n  
∂C(β) X exp (β0 + β1 xi )
=− yi xi − xi .
∂β1 i=1
1 + exp (β0 + β1 xi )

5
A more compact expression
Let us now define a vector y with n elements yi , an n × p matrix X which
contains the xi values and a vector p of fitted probabilities p(yi |xi , β). We can
rewrite in a more compact form the first derivative of cost function as

∂C(β)
= −X T (y − p) .
∂β
If we in addition define a diagonal matrix W with elements p(yi |xi , β)(1 −
p(yi |xi , β), we can obtain a compact expression of the second derivative as

∂ 2 C(β)
= X T W X.
∂β∂β T

Extending to more predictors

Within a binary classification problem, we can easily expand our model to include
multiple predictors. Our ratio between likelihoods is then with p predictors

p(βx)
log = β 0 + β1 x 1 + β2 x 2 + · · · + βp x p .
1 − p(βx)

Here we defined x = [1, x1 , x2 , . . . , xp ] and β = [β0 , β1 , . . . , βp ] leading to

exp (β0 + β1 x1 + β2 x2 + · · · + βp xp )
p(βx) = .
1 + exp (β0 + β1 x1 + β2 x2 + · · · + βp xp )

Including more classes

Till now we have mainly focused on two classes, the so-called binary system.
Suppose we wish to extend to K classes. Let us for the sake of simplicity assume
we have only two predictors. We have then following model

p(C = 1|x)
log = β10 + β11 x1 ,
p(K|x)
p(C = 2|x)
log = β20 + β21 x1 ,
p(K|x)
and so on till the class C = K − 1 class
p(C = K − 1|x)
log = β(K−1)0 + β(K−1)1 x1 ,
p(K|x)

and the model is specified in term of K − 1 so-called log-odds or logit

transformations.

6
More classes
In our discussion of neural networks we will encounter the above again in terms
of a slightly modified function, the so-called Softmax function.
The softmax function is used in various multiclass classification methods, such
as multinomial logistic regression (also known as softmax regression), multiclass
linear discriminant analysis, naive Bayes classifiers, and artificial neural networks.
Specifically, in multinomial logistic regression and linear discriminant analysis,
the input to the function is the result of K distinct linear functions, and the
predicted probability for the k-th class given a sample vector x and a weighting
vector β is (with two predictors):

exp (βk0 + βk1 x1 )

p(C = k|x) = PK−1 .
1 + l=1 exp (βl0 + βl1 x1 )
It is easy to extend to more predictors. The final class is
1
p(C = K|x) = PK−1 ,
1+ l=1 exp (βl0 + βl1 x1 )
and they sum to one. Our earlier discussions were all specialized to the case
with two classes only. It is easy to see from the above that what we derived
earlier is compatible with these equations.
To find the optimal parameters we would typically use a gradient descent
method. Newton’s method and gradient descent methods are discussed in the
material on optimization methods.

Preprocessing our data

We discuss here how to preprocess our data. Till now and in connection with
our previous examples we have not met so many cases where we are too sensitive
to the scaling of our data. Normally the data may need a rescaling and/or may
be sensitive to extreme values. Scaling the data renders our inputs much more
suitable for the algorithms we want to employ.
Scikit-Learn has several functions which allow us to rescale the data, nor-
mally resulting in much better results in terms of various accuracy scores.
The StandardScaler function in Scikit-Learn ensures that for each fea-
ture/predictor we study the mean value is zero and the variance is one (every
column in the design/feature matrix). This scaling has the drawback that it
does not ensure that we have a particular maximum or minimum in our data
set. Another function included in Scikit-Learn is the MinMaxScaler which
ensures that all features are exactly between 0 and 1. The

More preprocessing
The Normalizer scales each data point such that the feature vector has a
euclidean length of one. In other words, it projects a data point on the circle
(or sphere in the case of higher dimensions) with a radius of 1. This means

7
every data point is scaled by a different number (by the inverse of it’s length).
This normalization is often used when only the direction (or angle) of the data
matters, not the length of the feature vector.
The RobustScaler works similarly to the StandardScaler in that it ensures
statistical properties for each feature that guarantee that they are on the same
scale. However, the RobustScaler uses the median and quartiles, instead of mean
and variance. This makes the RobustScaler ignore data points that are very
different from the rest (like measurement errors). These odd data points are also
called outliers, and might often lead to trouble for other scaling techniques.

Simple preprocessing examples, breast cancer data and clas-

sification
We show here how we can use a simple regression case on the breast cancer data
using logistic regression as algorithm for classification.
import matplotlib.pyplot as plt
import numpy as np
from sklearn.model_selection import train_test_split
from sklearn.datasets import load_breast_cancer
from sklearn.linear_model import LogisticRegression
cancer = load_breast_cancer()

# Set up training data

X_train, X_test, y_train, y_test = train_test_split(cancer.data,cancer.target,random_state=0)
logreg = LogisticRegression()
logreg.fit(X_train, y_train)
print("Test set accuracy: {:.2f}".format(logreg.score(X_test,y_test)))
# Scale data
from sklearn.preprocessing import StandardScaler
scaler = StandardScaler()
scaler.fit(X_train)
X_train_scaled = scaler.transform(X_train)
X_test_scaled = scaler.transform(X_test)
logreg.fit(X_train_scaled, y_train)
print("Test set accuracy scaled data: {:.2f}".format(logreg.score(X_test_scaled,y_test)))

Covariance and Correlation

In addition to the plot of the features, we study now also the covariance (and
the correlation matrix). We use also Pandas to compute the correlation matrix.
import matplotlib.pyplot as plt
import numpy as np
from sklearn.model_selection import train_test_split
from sklearn.datasets import load_breast_cancer
from sklearn.linear_model import LogisticRegression
cancer = load_breast_cancer()
import pandas as pd
# Making a data frame
cancerpd = pd.DataFrame(cancer.data, columns=cancer.feature_names)

fig, axes = plt.subplots(15,2,figsize=(10,20))

malignant = cancer.data[cancer.target == 0]

8
 benign = cancer.data[cancer.target == 1]
ax = axes.ravel()

for i in range(30):
_, bins = np.histogram(cancer.data[:,i], bins =50)
ax[i].hist(malignant[:,i], bins = bins, alpha = 0.5)
ax[i].hist(benign[:,i], bins = bins, alpha = 0.5)
ax[i].set_title(cancer.feature_names[i])
ax[i].set_yticks(())
ax[0].set_xlabel("Feature magnitude")
ax[0].set_ylabel("Frequency")
ax[0].legend(["Malignant", "Benign"], loc ="best")
fig.tight_layout()
plt.show()

import seaborn as sns

sns.set(rc={’figure.figsize’:(15.0,15.0)},font_scale=1)
correlation_matrix = cancerpd.corr().round(1)
# use the heatmap function from seaborn to plot the correlation matrix
# annot = True to print the values inside the square
sns.heatmap(data=correlation_matrix, annot=True)
plt.show()

#print eigvalues of correlation matrix

EigValues, EigVectors = np.linalg.eig(correlation_matrix)
print(EigValues)

In the above example we note two things. In the first plot we display the
overlap of benign and malignant tumors as functions of the various features in
the Wisconsing breast cancer data set. We see that for some of the features we
can distinguish clearly the benign and malignant cases while for other features
we cannot. This can point to us which features may be of greater interest when
we wish to classify a benign or not benign tumour.
In the second figure we have computed the so-called correlation matrix, which
in our case with thirty features becomes a 30 × 30 matrix.
We constructed this matrix using pandas via the statements
cancerpd = pd.DataFrame(cancer.data, columns=cancer.feature_names)

and then
correlation_matrix = cancerpd.corr().round(1)

Diagonalizing this matrix we can in turn say something about which features
are of relevance and which are not. This leads us to the classical Principal
Component Analysis (PCA) theorem with applications. This topic is covered in
the PCA material and additional topics on dimensionality reduction.

Optimization, the central part of any Machine Learning

algortithm
Almost every problem in machine learning and data science starts with a dataset
X, a model g(β), which is a function of the parameters β and a cost function
C(X, g(β)) that allows us to judge how well the model g(β) explains the obser-
vations X. The model is fit by finding the values of β that minimize the cost

9
function. Ideally we would be able to solve for β analytically, however this is
not possible in general and we must use some approximative/numerical method
to compute the minimum.

Revisiting our Logistic Regression case

In our discussion on Logistic Regression we studied the case of two classes, with
yi either 0 or 1. Furthermore we assumed also that we have only two parameters
β in our fitting, that is we defined probabilities

exp (β0 + β1 xi )
p(yi = 1|xi , β) = ,
1 + exp (β0 + β1 xi )
p(yi = 0|xi , β) = 1 − p(yi = 1|xi , β),

where β are the weights we wish to extract from data, in our case β0 and β1 .

The equations to solve

Our compact equations used a definition of a vector y with n elements yi , an
n × p matrix X which contains the xi values and a vector p of fitted probabilities
p(yi |xi , β). We rewrote in a more compact form the first derivative of the cost
function as

∂C(β)
= −X T (y − p) .
∂β
If we in addition define a diagonal matrix W with elements p(yi |xi , β)(1 −
p(yi |xi , β), we can obtain a compact expression of the second derivative as

∂ 2 C(β)
= X T W X.
∂β∂β T
This defines what is called the Hessian matrix.

Solving using Newton-Raphson’s method

If we can set up these equations, Newton-Raphson’s iterative method is normally
the method of choice. It requires however that we can compute in an efficient
way the matrices that define the first and second derivatives.
Our iterative scheme is then given by
−1
∂ 2 C(β)
  
new old ∂C(β)
β =β − × ,
∂β∂β T β old ∂β β old

or in matrix form as

−1
β new = β old − X T W X × −X T (y − p)


β old
.

10
The right-hand side is computed with the old values of β.
If we can compute these matrices, in particular the Hessian, the above is
often the easiest method to implement.

Brief reminder on Newton-Raphson’s method

Let us quickly remind ourselves how we derive the above method.
Perhaps the most celebrated of all one-dimensional root-finding routines
is Newton’s method, also called the Newton-Raphson method. This method
requires the evaluation of both the function f and its derivative f 0 at arbitrary
points. If you can only calculate the derivative numerically and/or your function
is not of the smooth type, we normally discourage the use of this method.

The equations
The Newton-Raphson formula consists geometrically of extending the tangent
line at a current point until it crosses zero, then setting the next guess to the
abscissa of that zero-crossing. The mathematics behind this method is rather
simple. Employing a Taylor expansion for x sufficiently close to the solution s,
we have

(s − x)2 00
f (s) = 0 = f (x) + (s − x)f 0 (x) + f (x) + . . . .
2
For small enough values of the function and for well-behaved functions, the
terms beyond linear are unimportant, hence we obtain

f (x) + (s − x)f 0 (x) ≈ 0,

yielding
f (x)
s≈x− .
f 0 (x)
Having in mind an iterative procedure, it is natural to start iterating with
f (xn )
xn+1 = xn − .
f 0 (xn )

Simple geometric interpretation

The above is Newton-Raphson’s method. It has a simple geometric interpretation,
namely xn+1 is the point where the tangent from (xn , f (xn )) crosses the x-axis.
Close to the solution, Newton-Raphson converges fast to the desired result.
However, if we are far from a root, where the higher-order terms in the series
are important, the Newton-Raphson formula can give grossly inaccurate results.
For instance, the initial guess for the root might be so far from the true root as
to let the search interval include a local maximum or minimum of the function.
If an iteration places a trial guess near such a local extremum, so that the first
derivative nearly vanishes, then Newton-Raphson may fail totally

11
Extending to more than one variable
Newton’s method can be generalized to systems of several non-linear equations
and variables. Consider the case with two equations

f1 (x1 , x2 ) =0
f2 (x1 , x2 ) = 0,

which we Taylor expand to obtain

0 = f1 (x1 + h1 , x2 + h2 ) = f1 (x1 , x2 ) + h1 ∂f1 /∂x1 + h2 ∂f1 /∂x2 + . . .

.
0 = f2 (x1 + h1 , x2 + h2 ) = f2 (x1 , x2 ) + h1 ∂f2 /∂x1 + h2 ∂f2 /∂x2 + . . .

Defining the Jacobian matrix J we have

 
∂f1 /∂x1 ∂f1 /∂x2
J= ,
∂f2 /∂x1 ∂f2 /∂x2

we can rephrase Newton’s method as

 n+1   n   n 
x1 x1 h1
= + ,
xn+1
2 xn2 hn2

where we have defined

hn1 f1 (xn1 , xn2 )
   
= −J−1 .
hn2 f2 (xn1 , xn2 )

We need thus to compute the inverse of the Jacobian matrix and it is to

understand that difficulties may arise in case J is nearly singular.
It is rather straightforward to extend the above scheme to systems of more
than two non-linear equations. In our case, the Jacobian matrix is given by the
Hessian that represents the second derivative of cost function.

Steepest descent
The basic idea of gradient descent is that a function F (x), x ≡ (x1 , · · · , xn ),
decreases fastest if one goes from x in the direction of the negative gradient
−∇F (x).
It can be shown that if

xk+1 = xk − γk ∇F (xk ),

with γk > 0.
For γk small enough, then F (xk+1 ) ≤ F (xk ). This means that for a suffi-
ciently small γk we are always moving towards smaller function values, i.e a
minimum.

12
More on Steepest descent
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
x0 for a minimum of F and computes new approximations according to

xk+1 = xk − γk ∇F (xk ), k ≥ 0.
The parameter γk is often referred to as the step length or the learning rate
within the context of Machine Learning.

The ideal
Ideally the sequence {xk }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 minima. 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 x = (x1 , · · · , xn ) which makes it
expensive to compute numerically.

The sensitiveness of the gradient descent

The gradient descent method is sensitive to the choice of learning rate γk . This is
due to the fact that we are only guaranteed that F (xk+1 ) ≤ F (xk ) for sufficiently
small γk . The problem is to determine an optimal learning rate. If the learning
rate is chosen too small the method will take a long time to converge and if it is
too 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.

Convex functions
Ideally we want our cost/loss function to be convex(concave).
First we give the definition of a convex set: A set C in Rn is said to be convex
if, for all x and y in C and all t ∈ (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 R are the intervals of R. Examples of convex sets of
R2 are the regular polygons (triangles, rectangles, pentagons, etc...).

13
Convex function
Convex function: Let X ⊂ Rn be a convex set. Assume that the function
f : X → R is continuous, then f is said to be convex if

f (tx1 + (1 − t)x2 ) ≤ tf (x1 ) + (1 − t)f (x2 )

for all x1 , x2 ∈ X and for all t ∈ [0, 1]. If ≤ is replaced with a strict inequaltiy
in the definition, we demand x1 = 6 x2 and t ∈ (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 (x1 ) and f (x2 ), the value of the function on the interval
[x1 , x2 ] is always below the line as illustrated below.

Conditions on convex functions

In the following we state first and second-order conditions which ensures convexity
of a function f . We write Df to denote the domain of f , i.e the subset of Rn
where f is defined. For more details and proofs we refer to: S. Boyd and L.
Vandenberghe. Convex Optimization. Cambridge University Press.

First order condition. Suppose f is differentiable (i.e ∇f (x) is well defined

for all x in the domain of f ). Then f is convex if and only if Df is a convex set
and
f (y) ≥ f (x) + ∇f (x)T (y − x)
holds for all x, y ∈ Df . 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) = x2 + 1 and draw the tangent line to f (x) and note that it is always below
the graph.

Second order condition. Assume that f is twice differentiable, i.e the Hessian
matrix exists at each point in Df . Then f is convex if and only if Df is a convex
set and its Hessian is positive semi-definite for all x ∈ Df . For a single-variable
function this reduces to f 00 (x) ≥ 0. Geometrically this means that f has
nonnegative curvature everywhere.
This condition is particularly useful since it gives us an procedure for de-
termining if the function under consideration is convex, apart from using the
definition.

More on convex functions

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.

14
 Ideally we want the global minimum (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:

Any minimum is global for convex functions. Consider the problem of

finding x ∈ Rn such that f (x) is minimal, where f is convex and differentiable.
Then, any point x∗ that satisfies ∇f (x∗ ) = 0 is a global minimum.
This result means that if we know that the cost/loss function is convex and
we are able to find a minimum, we are guaranteed that it is a global minimum.

Some simple problems

1. Show that f (x) = x2 is convex for x ∈ R using the definition of convexity.
Hint: If you re-write the definition, f is convex if the following holds for all
x, y ∈ Df and any λ ∈ [0, 1] λf (x) + (1 − λ)f (y) − f (λx + (1 − λ)y) ≥ 0.

2. Using the second order condition show that the following functions are
convex on the specified domain.
• f (x) = ex is convex for x ∈ R.
• g(x) = − ln(x) is convex for x ∈ (0, ∞).

3. Let f (x) = x2 and g(x) = ex . Show that f (g(x)) and g(f (x)) is convex for
x ∈ R. Also show that if f (x) is any convex function than h(x) = ef (x) is
convex.
4. A norm is any function that satisfy the following properties
• f (αx) = |α|f (x) for all α ∈ R.
• f (x + y) ≤ f (x) + f (y)
• f (x) ≤ 0 for all x ∈ Rn with equality if and only if x = 0

Using the definition of convexity, try to show that a function satisfying the
properties above is convex (the third condition is not needed to show this).

Revisiting our first homework

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:

1. An analytical solution (recall homework set 1).

2. The gradient can be computed analytically.
3. The cost function is convex which guarantees that gradient descent con-
verges for small enough learning rates

15
We revisit the example from homework set 1 where we had

yi = 5x2i + 0.1ξi , i = 1, · · · , 100

with xi ∈ [0, 1] chosen randomly with a uniform distribution. Additionally ξi

represents stochastic noise chosen according to a normal distribution N (0, ∞).
The linear regression model is given by

hβ (x) = y = β0 + β1 x,

such that
yi = β0 + β1 xi .

Gradient descent example

Let y = (y1 , · · · , yn )T , y = (y1 , · · · , yn )T and β = (β0 , β1 )T
It is convenient to write y = Xβ where X ∈ R100×2 is the design matrix
given by  
1 amp; x1
X ≡  ... .
amp; .. .
 

1 amp; x100 amp;

The loss function is given by
100
X
C(β) = ||Xβ−y||2 = ||Xβ||2 −2yT Xβ+||y||2 = (β0 +β1 xi )2 −2yi (β0 +β1 xi )+yi2
i=1

and we want to find β such that C(β) is minimized.

The derivative of the cost/loss function

Computing ∂C(β)/∂β0 and ∂C(β)/∂β1 we can show that the gradient can be
written as
 P100 
(β + β1 xi − yi )
∇β C(β) = (∂C(β)/∂β0 , ∂C(β)/∂β1 )T = 2 P100 i=1 0 = 2X T (Xβ−y),
i=1 (xi (β 0 + β 1 x i ) − yi x i )
where X is the design matrix defined above.

The Hessian matrix

The Hessian matrix of C(β) is given by
 2 
∂ C(β) ∂ 2 C(β)
∂β 2 amp; ∂β 0 ∂β1
H ≡  ∂ 2 C(β)
0  = 2X T X.
∂ 2 C(β)
∂β0 ∂β1 amp; ∂β 2 amp;
1

This result implies that C(β) is a convex function since the matrix X T X always
is positive semi-definite.

16
Simple program
We can now write a program that minimizes C(β) using the gradient descent
method with a constant learning rate γ according to

βk+1 = βk − γ∇β C(βk ), k = 0, 1, · · ·

We can use the expression we computed for the gradient and let use a β0 be
chosen randomly and let γ = 0.001. Stop iterating when ||∇β C(βk )|| ≤  = 10−8 .
And finally we can compare our solution for β with the analytic result given
by β = (X T X)−1 X T y.

Gradient Descent Example

Here our simple example
# 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

# the number of datapoints

m = 100
x = 2*np.random.rand(m,1)
y = 4+3*x+np.random.randn(m,1)
xb = np.c_[np.ones((m,1)), x]
beta_linreg = np.linalg.inv(xb.T.dot(xb)).dot(xb.T).dot(y)
print(beta_linreg)
beta = np.random.randn(2,1)
eta = 0.1
Niterations = 1000
for iter in range(Niterations):
gradients = 2.0/m*xb.T.dot(xb.dot(beta)-y)
beta -= eta*gradients

print(beta)
xnew = np.array([[0],[2]])
xbnew = np.c_[np.ones((2,1)), xnew]
ypredict = xbnew.dot(beta)
ypredict2 = xbnew.dot(beta_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’Gradient descent example’)
plt.show()

17
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]
beta_linreg = np.linalg.inv(xb.T.dot(xb)).dot(xb.T).dot(y)
print(beta_linreg)
sgdreg = SGDRegressor(n_iter = 50, penalty=None, eta0=0.1)
sgdreg.fit(x,y.ravel())
print(sgdreg.intercept_, sgdreg.coef_)

Gradient descent and Ridge

We have also discussed Ridge regression where the loss function contains a
regularized term given by the L2 norm of β,

Cridge (β) = ||Xβ − y||2 + λ||β||2 , λ ≥ 0.

In order to minimize Cridge (β) using GD we only have adjust the gradient as
follows
 P100   
(β0 + β1 xi − yi ) β
∇β Cridge (β) = 2 100
i=1 +2λ 0 = 2(X T (Xβ−y)+λβ).
β1
P
i=1 (x (β
i 0 + β x
1 i ) − y x
i i )

We can easily extend our program to minimize Cridge (β) using gradient
descent and compare with the analytical solution given by

−1
βridge = X T X + λI2×2 X T y.

Program example for gradient descent with Ridge Regres-

sion
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

# the number of datapoints

m = 100
x = 2*np.random.rand(m,1)
y = 4+3*x+np.random.randn(m,1)
xb = np.c_[np.ones((m,1)), x]
XT_X = xb.T @ xb

18
 #Ridge parameter lambda
lmbda = 0.001
Id = lmbda* np.eye(XT_X.shape[0])

beta_linreg = np.linalg.inv(XT_X+Id) @ xb.T @ y

print(beta_linreg)
# Start plain gradient descent
beta = np.random.randn(2,1)

eta = 0.1
Niterations = 100

for iter in range(Niterations):

gradients = 2.0/m*xb.T @ (xb @ (beta)-y)+2*lmbda*beta
beta -= eta*gradients

print(beta)
ypredict = xb @ beta
ypredict2 = xb @ beta_linreg
plt.plot(x, ypredict, "r-")
plt.plot(x, 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’Gradient descent example for Ridge’)
plt.show()

Using gradient descent methods, limitations

• Gradient descent (GD) finds local minima of our function. Since
the GD algorithm is deterministic, if it converges, it will converge to a
local minimum of our energy function. Because in ML we are often dealing
with extremely rugged landscapes with many local minima, this can lead
to poor performance.
• GD is sensitive to initial conditions. One consequence of the local
nature of GD is that initial conditions matter. Depending on where one
starts, one will end up at a different local minima. Therefore, it is very
important to think about how one initializes the training process. This is
true for GD as well as more complicated variants of GD.
• Gradients are computationally expensive to calculate for large
datasets. In many cases in statistics and ML, the energy function is a
sum of terms, with
Pn one term for each data point. For example, in linear
regression, E ∝ i=1 (yi − wT · xi )2 ; for logistic regression, the square error
is replaced by the cross entropy. To calculate the gradient we have to sum
over all n data points. Doing this at every GD step becomes extremely
computationally expensive. An ingenious solution to this, is to calculate
the gradients using small subsets of the data called “mini batches”. This
has the added benefit of introducing stochasticity into our algorithm.

19
 • GD is very sensitive to choices of learning rates. GD is extremely
sensitive to the choice of learning rates. If the learning rate is very small,
the training process take an extremely long time. For larger learning rates,
GD can diverge and give poor results. Furthermore, depending on what
the local landscape looks like, we have to modify the learning rates to
ensure convergence. Ideally, we would adaptively choose the learning rates
to match the landscape.
• GD treats all directions in parameter space uniformly. Another
major drawback of GD is that unlike Newton’s method, the learning rate
for GD is the same in all directions in parameter space. For this reason,
the maximum learning rate is set by the behavior of the steepest direction
and this can significantly slow down training. Ideally, we would like to
take large steps in flat directions and small steps in steep directions. Since
we are exploring rugged landscapes where curvatures change, this requires
us to keep track of not only the gradient but second derivatives. The
ideal scenario would be to calculate the Hessian but this proves to be too
computationally expensive.
• GD can take exponential time to escape saddle points, even with random
initialization. As we mentioned, GD is extremely sensitive to initial
condition since it determines the particular local minimum GD would
eventually reach. However, even with a good initialization scheme, through
the introduction of randomness, GD can still take exponential time to
escape saddle points.

Stochastic Gradient Descent

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 data
points {xi }ni=1 ,
Xn
C(β) = ci (xi , β).
i=1

Computation of gradients
This in turn means that the gradient can be computed as a sum over i-gradients
n
X
∇β C(β) = ∇β ci (xi , β).
i

Stochasticity/randomness is introduced by only taking the gradient on a

subset of the data called minibatches. If there are n data points and the size
of each minibatch is M , there will be n/M minibatches. We denote these
minibatches by Bk where k = 1, · · · , n/M .

20
SGD example
As an example, suppose we have 10 data points (x1 , · · · , x10 ) and we choose
to have M = 5 minibathces, then each minibatch contains two data points. In
particular we have B1 = (x1 , x2 ), · · · , B5 = (x9 , x10 ). Note that if you choose
M = 1 you have only a single batch with all data points and on the other
extreme, you may choose M = n resulting in a minibatch for each datapoint, i.e
Bk = xk .
The idea is now to approximate the gradient by replacing the sum over all
data points with a sum over the data points in one the minibatches picked at
random in each gradient descent step
n
X n
X
∇β C(β) = ∇β ci (xi , β) → ∇β ci (xi , β).
i=1 i∈Bk

The gradient step

Thus a gradient descent step now looks like
n
X
βj+1 = βj − γj ∇β ci (xi , β)
i∈Bk

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.

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
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.

21
When do we stop?
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 β that gave the lowest value.

Slightly different approach

Another approach is to let the step length γj 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, · · · denote the current epoch and let t0 , t1 > 0
be two fixed numbers. Furthermore, let t = e · m + i where m is the number of
minibatches and i = 0, · · · , m − 1. Then the function
t0
γj (t; t0 , t1 ) =
t + t1
goes to zero as the number of epochs gets large. I.e. we start with a step length
γj (0; t0 , t1 ) = t0 /t1 which decays in time t.
In this way we can fix the number of epochs, compute β 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 β 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 beta
t = epoch*m+i
gamma_j = step_length(t,t0,t1)
j += 1

print("gamma_j after %d epochs: %g" % (n_epochs,gamma_j))

22
Program for stochastic gradient
# 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

m = 100
x = 2*np.random.rand(m,1)
y = 4+3*x+np.random.randn(m,1)

xb = np.c_[np.ones((m,1)), x]
theta_linreg = np.linalg.inv(xb.T.dot(xb)).dot(xb.T).dot(y)
print("Own inversion")
print(theta_linreg)
sgdreg = SGDRegressor(max_iter = 50, penalty=None, eta0=0.1)
sgdreg.fit(x,y.ravel())
print("sgdreg from scikit")
print(sgdreg.intercept_, sgdreg.coef_)

theta = np.random.randn(2,1)
eta = 0.1
Niterations = 1000

for iter in range(Niterations):

gradients = 2.0/m*xb.T @ ((xb @ theta)-y)
theta -= eta*gradients
print("theta frm own gd")
print(theta)

xnew = np.array([[0],[2]])
xbnew = np.c_[np.ones((2,1)), xnew]
ypredict = xbnew.dot(theta)
ypredict2 = xbnew.dot(theta_linreg)

n_epochs = 50
t0, t1 = 5, 50
def learning_schedule(t):
return t0/(t+t1)

theta = np.random.randn(2,1)

for epoch in range(n_epochs):

for i in range(m):
random_index = np.random.randint(m)
xi = xb[random_index:random_index+1]
yi = y[random_index:random_index+1]
gradients = 2 * xi.T @ ((xi @ theta)-yi)
eta = learning_schedule(epoch*m+i)
theta = theta - eta*gradients
print("theta from own sdg")
print(theta)

plt.plot(xnew, ypredict, "r-")

plt.plot(xnew, ypredict2, "b-")
plt.plot(x, y ,’ro’)

23
 plt.axis([0,2.0,0, 15.0])
plt.xlabel(r’$x$’)
plt.ylabel(r’$y$’)
plt.title(r’Random numbers ’)
plt.show()

Challenge: try to write a similar code for a Logistic Regression case.

Momentum based GD
The stochastic gradient descent (SGD) is almost always used with a momentum
or inertia term that serves as a memory of the direction we are moving in
parameter space. This is typically implemented as follows

vt = γvt−1 + ηt ∇θ E(θt )
θt+1 = θt − vt , (2)

where we have introduced a momentum parameter γ, with 0 ≤ γ ≤ 1, and for

brevity we dropped the explicit notation to indicate the gradient is to be taken
over a different mini-batch at each step. We call this algorithm gradient descent
with momentum (GDM). From these equations, it is clear that vt is a running
average of recently encountered gradients and (1 − γ)−1 sets the characteristic
time scale for the memory used in the averaging procedure. Consistent with this,
when γ = 0, this just reduces down to ordinary SGD as discussed earlier. An
equivalent way of writing the updates is

∆θt+1 = γ∆θt − ηt ∇θ E(θt ),

where we have defined ∆θt = θt − θt−1 .

More on momentum based approaches

Let us try to get more intuition from these equations. It is helpful to consider a
simple physical analogy with a particle of mass m moving in a viscous medium
with drag coefficient µ and potential E(w). If we denote the particle’s position
by w, then its motion is described by

d2 w dw
m +µ = −∇w E(w).
dt2 dt
We can discretize this equation in the usual way to get
wt+∆t − 2wt + wt−∆t wt+∆t − wt
m +µ = −∇w E(w).
(∆t)2 ∆t
Rearranging this equation, we can rewrite this as

(∆t)2 m
∆wt+∆t = − ∇w E(w) + ∆wt .
m + µ∆t m + µ∆t

24
Momentum parameter
Notice that this equation is identical to previous one if we identify the position of
the particle, w, with the parameters θ. This allows us to identify the momentum
parameter and learning rate with the mass of the particle and the viscous drag
as:

m (∆t)2
γ= , η= .
m + µ∆t m + µ∆t
Thus, as the name suggests, the momentum parameter is proportional to
the mass of the particle and effectively provides inertia. Furthermore, in the
large viscosity/small learning rate limit, our memory time scales as (1 − γ)−1 ≈
m/(µ∆t).
Why is momentum useful? SGD momentum helps the gradient descent
algorithm gain speed in directions with persistent but small gradients even in
the presence of stochasticity, while suppressing oscillations in high-curvature
directions. This becomes especially important in situations where the landscape
is shallow and flat in some directions and narrow and steep in others. It has
been argued that first-order methods (with appropriate initial conditions) can
perform comparable to more expensive second order methods, especially in the
context of complex deep learning models.
These beneficial properties of momentum can sometimes become even more
pronounced by using a slight modification of the classical momentum algorithm
called Nesterov Accelerated Gradient (NAG).
In the NAG algorithm, rather than calculating the gradient at the current
parameters, ∇θ E(θt ), one calculates the gradient at the expected value of the
parameters given our current momentum, ∇θ E(θt + γvt−1 ). This yields the
NAG update rule

vt = γvt−1 + ηt ∇θ E(θt + γvt−1 )

θt+1 = θt − vt . (3)

One of the major advantages of NAG is that it allows for the use of a larger
learning rate than GDM for the same choice of γ.

Second moment of the gradient

In stochastic gradient descent, with and without momentum, we still have to
specify a schedule for tuning the learning rates ηt as a function of time. As
discussed in the context of Newton’s method, this presents a number of dilemmas.
The learning rate is limited by the steepest direction which can change depending
on the current position in the landscape. To circumvent this problem, ideally
our algorithm would keep track of curvature and take large steps in shallow, flat
directions and small steps in steep, narrow directions. Second-order methods
accomplish this by calculating or approximating the Hessian and normalizing the

25
learning rate by the curvature. However, this is very computationally expensive
for extremely large models. Ideally, we would like to be able to adaptively change
the step size to match the landscape without paying the steep computational
price of calculating or approximating Hessians.
Recently, a number of methods have been introduced that accomplish this
by tracking not only the gradient, but also the second moment of the gradient.
These methods include AdaGrad, AdaDelta, RMS-Prop, and ADAM.

RMS prop
In RMS prop, in addition to keeping a running average of the first moment of
the gradient, we also keep track of the second moment denoted by st = E[gt2 ].
The update rule for RMS prop is given by

gt = ∇θ E(θ) (4)
st = βst−1 + (1 − β)gt2
gt
θt+1 = θt − η t √ ,
st + 

where β controls the averaging time of the second moment and is typically
taken to be about β = 0.9, ηt is a learning rate typically chosen to be 10−3 , and
 ∼ 10−8 is a small regularization constant to prevent divergences. Multiplication
and division by vectors is understood as an element-wise operation. It is clear
from this formula that the learning rate is reduced in directions where the norm
of the gradient is consistently large. This greatly speeds up the convergence by
allowing us to use a larger learning rate for flat directions.

ADAM optimizer
A related algorithm is the ADAM optimizer. In ADAM, we keep a running
average of both the first and second moment of the gradient and use this
information to adaptively change the learning rate for different parameters. In
addition to keeping a running average of the first and second moments of the
gradient (i.e. mt = E[gt ] and st = E[gt2 ], respectively), ADAM performs an
additional bias correction to account for the fact that we are estimating the first
two moments of the gradient using a running average (denoted by the hats in the
update rule below). The update rule for ADAM is given by (where multiplication
and division are once again understood to be element-wise operations below)

26
 gt = ∇θ E(θ) (5)
mt = β1 mt−1 + (1 − β1 )gt
st = β2 st−1 + (1 − β2 )gt2
mt
mt =
1 − β1t
st
st =
1 − β2t
mt
θt+1 = θt − ηt √ ,
st + 
(6)

where β1 and β2 set the memory lifetime of the first and second moment and
are typically taken to be 0.9 and 0.99 respectively, and η and  are identical to
RMSprop.
Like in RMSprop, the effective step size of a parameter depends on the
magnitude of its gradient squared. To understand this better, let us rewrite this
expression in terms of the variance σt2 = st − (mt )2 . Consider a single parameter
θt . The update rule for this parameter is given by
mt
∆θt+1 = −ηt p .
σt2 + m2t + 

Practical tips
• Randomize the data when making mini-batches. It is always impor-
tant to randomly shuffle the data when forming mini-batches. Otherwise,
the gradient descent method can fit spurious correlations resulting from
the order in which data is presented.
• Transform your inputs. Learning becomes difficult when our landscape
has a mixture of steep and flat directions. One simple trick for minimizing
these situations is to standardize the data by subtracting the mean and
normalizing the variance of input variables. Whenever possible, also
decorrelate the inputs. To understand why this is helpful, consider the
case of linear regression. It is easy to show that for the squared error cost
function, the Hessian of the energy matrix is just the correlation matrix
between the inputs. Thus, by standardizing the inputs, we are ensuring
that the landscape looks homogeneous in all directions in parameter space.
Since most deep networks can be viewed as linear transformations followed
by a non-linearity at each layer, we expect this intuition to hold beyond
the linear case.
• Monitor the out-of-sample performance. Always monitor the perfor-
mance of your model on a validation set (a small portion of the training

27
 data that is held out of the training process to serve as a proxy for the test
set. If the validation error starts increasing, then the model is beginning to
overfit. Terminate the learning process. This early stopping significantly
improves performance in many settings.
• Adaptive optimization methods don’t always have good general-
ization. Recent studies have shown that adaptive methods such as ADAM,
RMSPorp, and AdaGrad tend to have poor generalization compared to
SGD or SGD with momentum, particularly in the high-dimensional limit
(i.e. the number of parameters exceeds the number of data points). Al-
though it is not clear at this stage why these methods perform so well
in training deep neural networks, simpler procedures like properly-tuned
SGD may work as well or better in these applications.

Geron’s text, see chapter 11, has several interesting discussions.

Automatic differentiation
Automatic differentiation (AD), also called algorithmic differentiation or com-
putational differentiation,is a set of techniques to numerically evaluate the
derivative of a function specified by a computer program. AD exploits the fact
that every computer program, no matter how complicated, executes a sequence
of elementary arithmetic operations (addition, subtraction, multiplication, divi-
sion, etc.) and elementary functions (exp, log, sin, cos, etc.). By applying the
chain rule repeatedly to these operations, derivatives of arbitrary order can be
computed automatically, accurately to working precision, and using at most a
small constant factor more arithmetic operations than the original program.
Automatic differentiation is neither:

• Symbolic differentiation, nor

• Numerical differentiation (the method of finite differences).

Symbolic differentiation can lead to inefficient code and faces the difficulty
of converting a computer program into a single expression, while numerical
differentiation can introduce round-off errors in the discretization process and
cancellation
Python has tools for so-called automatic differentiation. Consider the
following example
f (x) = sin 2πx + x2

which has the following derivative

f 0 (x) = cos 2πx + x2 (2π + 2x)

Using autograd we have

28
 import autograd.numpy as np
# To do elementwise differentiation:
from autograd import elementwise_grad as egrad

# To plot:
import matplotlib.pyplot as plt

def f(x):
return np.sin(2*np.pi*x + x**2)

def f_grad_analytic(x):
return np.cos(2*np.pi*x + x**2)*(2*np.pi + 2*x)
# Do the comparison:
x = np.linspace(0,1,1000)
f_grad = egrad(f)

computed = f_grad(x)
analytic = f_grad_analytic(x)

plt.title(’Derivative computed from Autograd compared with the analytical derivative’)

plt.plot(x,computed,label=’autograd’)
plt.plot(x,analytic,label=’analytic’)

plt.xlabel(’x’)
plt.ylabel(’y’)
plt.legend()

plt.show()

print("The max absolute difference is: %g"%(np.max(np.abs(computed - analytic))))

Using autograd
Here we experiment with what kind of functions Autograd is capable of finding
the gradient of. The following Python functions are just meant to illustrate what
Autograd can do, but please feel free to experiment with other, possibly more
complicated, functions as well.
import autograd.numpy as np
from autograd import grad
def f1(x):
return x**3 + 1
f1_grad = grad(f1)

# Remember to send in float as argument to the computed gradient from Autograd!

a = 1.0
# See the evaluated gradient at a using autograd:
print("The gradient of f1 evaluated at a = %g using autograd is: %g"%(a,f1_grad(a)))
# Compare with the analytical derivative, that is f1’(x) = 3*x**2
grad_analytical = 3*a**2
print("The gradient of f1 evaluated at a = %g by finding the analytic expression is: %g"%(a,grad_a

29
Autograd with more complicated functions
To differentiate with respect to two (or more) arguments of a Python function,
Autograd need to know at which variable the function if being differentiated
with respect to.
import autograd.numpy as np
from autograd import grad
def f2(x1,x2):
return 3*x1**3 + x2*(x1 - 5) + 1

# By sending the argument 0, Autograd will compute the derivative w.r.t the first variable, in thi
f2_grad_x1 = grad(f2,0)

# ... and differentiate w.r.t x2 by sending 1 as an additional arugment to grad

f2_grad_x2 = grad(f2,1)

x1 = 1.0
x2 = 3.0

print("Evaluating at x1 = %g, x2 = %g"%(x1,x2))

print("-"*30)

# Compare with the analytical derivatives:

# Derivative of f2 w.r.t x1 is: 9*x1**2 + x2:

f2_grad_x1_analytical = 9*x1**2 + x2

# Derivative of f2 w.r.t x2 is: x1 - 5:

f2_grad_x2_analytical = x1 - 5

# See the evaluated derivations:

print("The derivative of f2 w.r.t x1: %g"%( f2_grad_x1(x1,x2) ))
print("The analytical derivative of f2 w.r.t x1: %g"%( f2_grad_x1(x1,x2) ))

print()
print("The derivative of f2 w.r.t x2: %g"%( f2_grad_x2(x1,x2) ))
print("The analytical derivative of f2 w.r.t x2: %g"%( f2_grad_x2(x1,x2) ))

Note that the grad function will not produce the true gradient of the function.
The true gradient of a function with two or more variables will produce a vector,
where each element is the function differentiated w.r.t a variable.

More complicated functions using the elements of their ar-

guments directly
import autograd.numpy as np
from autograd import grad
def f3(x): # Assumes x is an array of length 5 or higher
return 2*x[0] + 3*x[1] + 5*x[2] + 7*x[3] + 11*x[4]**2
f3_grad = grad(f3)

x = np.linspace(0,4,5)

# Print the computed gradient:

print("The computed gradient of f3 is: ", f3_grad(x))

30
 # The analytical gradient is: (2, 3, 5, 7, 22*x[4])
f3_grad_analytical = np.array([2, 3, 5, 7, 22*x[4]])

# Print the analytical gradient:

print("The analytical gradient of f3 is: ", f3_grad_analytical)

Note that in this case, when sending an array as input argument, the output
from Autograd is another array. This is the true gradient of the function, as
opposed to the function in the previous example. By using arrays to represent
the variables, the output from Autograd might be easier to work with, as the
output is closer to what one could expect form a gradient-evaluting function.

Functions using mathematical functions from Numpy

import autograd.numpy as np
from autograd import grad
def f4(x):
return np.sqrt(1+x**2) + np.exp(x) + np.sin(2*np.pi*x)

f4_grad = grad(f4)
x = 2.7

# Print the computed derivative:

print("The computed derivative of f4 at x = %g is: %g"%(x,f4_grad(x)))
# The analytical derivative is: x/sqrt(1 + x**2) + exp(x) + cos(2*pi*x)*2*pi
f4_grad_analytical = x/np.sqrt(1 + x**2) + np.exp(x) + np.cos(2*np.pi*x)*2*np.pi
# Print the analytical gradient:
print("The analytical gradient of f4 at x = %g is: %g"%(x,f4_grad_analytical))

More autograd
import autograd.numpy as np
from autograd import grad
def f5(x):
if x >= 0:
return x**2
else:
return -3*x + 1

f5_grad = grad(f5)
x = 2.7

# Print the computed derivative:

print("The computed derivative of f5 at x = %g is: %g"%(x,f5_grad(x)))

And with loops

import autograd.numpy as np
from autograd import grad
def f6_for(x):
val = 0

31
 for i in range(10):
val = val + x**i
return val

def f6_while(x):
val = 0
i = 0
while i < 10:
val = val + x**i
i = i + 1
return val

f6_for_grad = grad(f6_for)
f6_while_grad = grad(f6_while)

x = 0.5

# Print the computed derivaties of f6_for and f6_while

print("The computed derivative of f6_for at x = %g is: %g"%(x,f6_for_grad(x)))
print("The computed derivative of f6_while at x = %g is: %g"%(x,f6_while_grad(x)))

import autograd.numpy as np
from autograd import grad
# Both of the functions are implementation of the sum: sum(x**i) for i = 0, ..., 9
# The analytical derivative is: sum(i*x**(i-1))
f6_grad_analytical = 0
for i in range(10):
f6_grad_analytical += i*x**(i-1)

print("The analytical derivative of f6 at x = %g is: %g"%(x,f6_grad_analytical))

Using recursion
import autograd.numpy as np
from autograd import grad

def f7(n): # Assume that n is an integer

if n == 1 or n == 0:
return 1
else:
return n*f7(n-1)

f7_grad = grad(f7)

n = 2.0
print("The computed derivative of f7 at n = %d is: %g"%(n,f7_grad(n)))

# The function f7 is an implementation of the factorial of n.

# By using the product rule, one can find that the derivative is:

f7_grad_analytical = 0
for i in range(int(n)-1):
tmp = 1
for k in range(int(n)-1):
if k != i:
tmp *= (n - k)
f7_grad_analytical += tmp

print("The analytical derivative of f7 at n = %d is: %g"%(n,f7_grad_analytical))

32
Note that if n is equal to zero or one, Autograd will give an error message. This
message appears when the output is independent on input.

Unsupported functions
Autograd supports many features. However, there are some functions that is not
supported (yet) by Autograd.
Assigning a value to the variable being differentiated with respect to
import autograd.numpy as np
from autograd import grad
def f8(x): # Assume x is an array
x[2] = 3
return x*2

f8_grad = grad(f8)

x = 8.4
print("The derivative of f8 is:",f8_grad(x))

Here, Autograd tells us that an ’ArrayBox’ does not support item assignment.
The item assignment is done when the program tries to assign x[2] to the value
3. However, Autograd has implemented the computation of the derivative such
that this assignment is not possible.

The syntax a.dot(b) when finding the dot product

import autograd.numpy as np
from autograd import grad
def f9(a): # Assume a is an array with 2 elements
b = np.array([1.0,2.0])
return a.dot(b)
f9_grad = grad(f9)

x = np.array([1.0,0.0])

print("The derivative of f9 is:",f9_grad(x))

Here we are told that the ’dot’ function does not belong to Autograd’s version
of a Numpy array. To overcome this, an alternative syntax which also computed
the dot product can be used:
import autograd.numpy as np
from autograd import grad
def f9_alternative(x): # Assume a is an array with 2 elements
b = np.array([1.0,2.0])
return np.dot(x,b) # The same as x_1*b_1 + x_2*b_2
f9_alternative_grad = grad(f9_alternative)

x = np.array([3.0,0.0])

print("The gradient of f9 is:",f9_alternative_grad(x))

33
 # The analytical gradient of the dot product of vectors x and b with two elements (x_1,x_2) and (b
# w.r.t x is (b_1, b_2).

Recommended to avoid
The documentation recommends to avoid inplace operations such as
a += b
a -= b
a*= b
a /=b

Standard steepest descent

Before we proceed, we would like to discuss the approach called the standard
Steepest descent, which again leads to us having to be able to compute a
matrix. It belongs to the class of Conjugate Gradient methods (CG).
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
Ax = b.
In the iterative process we end up with a problem like

r = b − Ax,
where r is the so-called residual or error in the iterative process.
When we have found the exact solution, r = 0.

Gradient method
The residual is zero when we reach the minimum of the quadratic equation
1 T
P (x) = x Ax − xT b,
2
with the constraint that the matrix A is positive definite and symmetric.
This defines also the Hessian and we want it to be positive definite.

Steepest descent method

We denote the initial guess for x as x0 . We can assume without loss of generality
that
x0 = 0,
or consider the system
Az = b − Ax0 ,
instead.

34
Steepest descent method
One can show that the solution x is also the unique minimizer of the quadratic
form
1
f (x) = xT Ax − xT x, x ∈ Rn .
2
This suggests taking the first basis vector r1 (see below for definition) to be the
gradient of f at x = x0 , which equals

Ax0 − b,

and x0 = 0 it is equal −b.

Final expressions
We can compute the residual iteratively as

rk+1 = b − Axk+1 ,

which equals
b − A(xk + αk rk ),
or
(b − Axk ) − αk Ark ,
which gives

rkT rk
αk =
rkT Ark
leading to the iterative scheme

xk+1 = xk − αk rk ,

Code examples for steepest descent

Simple codes for steepest descent and conjugate gradient
using a 2 × 2 matrix, in c++, Python code to come

#include <cmath>
#include <iostream>
#include <fstream>
#include <iomanip>
#include "vectormatrixclass.h"
using namespace std;
// Main function begins here
int main(int argc, char * argv[]){
int dim = 2;
Vector x(dim),xsd(dim), b(dim),x0(dim);
Matrix A(dim,dim);

35
 // Set our initial guess
x0(0) = x0(1) = 0;
// Set the matrix
A(0,0) = 3; A(1,0) = 2; A(0,1) = 2; A(1,1) = 6;
b(0) = 2; b(1) = -8;
cout << "The Matrix A that we are using: " << endl;
A.Print();
cout << endl;
xsd = SteepestDescent(A,b,x0);
cout << "The approximate solution using Steepest Descent is: " << endl;
xsd.Print();
cout << endl;
}

The routine for the steepest descent method

Vector SteepestDescent(Matrix A, Vector b, Vector x0){

int IterMax, i;
int dim = x0.Dimension();
const double tolerance = 1.0e-14;
Vector x(dim),f(dim),z(dim);
double c,alpha,d;
IterMax = 30;
x = x0;
r = A*x-b;
i = 0;
while (i <= IterMax){
z = A*r;
c = dot(r,r);
alpha = c/dot(r,z);
x = x - alpha*r;
r = A*x-b;
if(sqrt(dot(r,r)) < tolerance) break;
i++;
}
return x;
}

Steepest descent example

import numpy as np
import numpy.linalg as la
import scipy.optimize as sopt

import matplotlib.pyplot as pt
from mpl_toolkits.mplot3d import axes3d
def f(x):
return 0.5*x[0]**2 + 2.5*x[1]**2
def df(x):
return np.array([x[0], 5*x[1]])

fig = pt.figure()
ax = fig.gca(projection="3d")

36
 xmesh, ymesh = np.mgrid[-2:2:50j,-2:2:50j]
fmesh = f(np.array([xmesh, ymesh]))
ax.plot_surface(xmesh, ymesh, fmesh)

And then as countor plot

pt.axis("equal")
pt.contour(xmesh, ymesh, fmesh)
guesses = [np.array([2, 2./5])]

Find guesses
x = guesses[-1]
s = -df(x)

Run it!
def f1d(alpha):
return f(x + alpha*s)

alpha_opt = sopt.golden(f1d)
next_guess = x + alpha_opt * s
guesses.append(next_guess)
print(next_guess)

What happened?
pt.axis("equal")
pt.contour(xmesh, ymesh, fmesh, 50)
it_array = np.array(guesses)
pt.plot(it_array.T[0], it_array.T[1], "x-")

Conjugate gradient method

In the CG method we define so-called conjugate directions and two vectors s
and t are said to be conjugate if

sT At = 0.

The philosophy of the CG method is to perform searches in various conjugate

directions of our vectors xi obeying the above criterion, namely

xTi Axj = 0.

Two vectors are conjugate if they are orthogonal with respect to this inner
product. Being conjugate is a symmetric relation: if s is conjugate to t, then t
is conjugate to s.

Conjugate gradient method

An example is given by the eigenvectors of the matrix

viT Avj = λviT vj ,

which is zero unless i = j.

37
Conjugate gradient method
Assume now that we have a symmetric positive-definite matrix A of size n × n.
At each iteration i + 1 we obtain the conjugate direction of a vector

xi+1 = xi + αi pi .

We assume that pi is a sequence of n mutually conjugate directions. Then the pi

form a basis of Rn and we can expand the solution Ax = b in this basis, namely
n
X
x= α i pi .
i=1

Conjugate gradient method

The coefficients are given by
n
X
Ax = αi Api = b.
i=1

Multiplying with pTk from the left gives

n
X
pTk Ax = αi pTk Api = pTk b,
i=1

and we can define the coefficients αk as

pTk b
αk =
pTk Apk

Conjugate gradient method and iterations

If we choose the conjugate vectors pk carefully, then we may not need all of
them to obtain a good approximation to the solution 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 x as x0 . We can assume without loss of
generality that
x0 = 0,
or consider the system
Az = b − Ax0 ,
instead.

38
Conjugate gradient method
One can show that the solution x is also the unique minimizer of the quadratic
form
1
f (x) = xT Ax − xT x, x ∈ Rn .
2
This suggests taking the first basis vector p1 to be the gradient of f at x = x0 ,
which equals
Ax0 − b,
and x0 = 0 it is equal −b. The other vectors in the basis will be conjugate to
the gradient, hence the name conjugate gradient method.

Conjugate gradient method

Let rk be the residual at the k-th step:

rk = b − Axk .

Note that rk is the negative gradient of f at x = xk , so the gradient descent

method would be to move in the direction rk . Here, we insist that the directions
pk are conjugate to each other, so we take the direction closest to the gradient
rk under the conjugacy constraint. This gives the following expression

pTk Ark
pk+1 = rk − pk .
pTk Apk

Conjugate gradient method

We can also compute the residual iteratively as

rk+1 = b − Axk+1 ,

which equals
b − A(xk + αk pk ),
or
(b − Axk ) − αk Apk ,
which gives

rk+1 = rk − Apk ,

Broyden–Fletcher–Goldfarb–Shanno algorithm
The optimization problem is to minimize f (x) where x is a vector in Rn , and
f is a differentiable scalar function. There are no constraints on the values that
x can take.
The algorithm begins at an initial estimate for the optimal value x0 and
proceeds iteratively to get a better estimate at each stage.

39
 The search direction pk at stage k is given by the solution of the analogue of
the Newton equation
Bk pk = −∇f (xk ),
where Bk is an approximation to the Hessian matrix, which is updated
iteratively at each stage, and ∇f (xk ) is the gradient of the function evaluated
at xk . A line search in the direction pk is then used to find the next point xk+1
by minimising
f (xk + αpk ),
over the scalar α > 0.

40