README.md

Classifier

Python, NumPy and Matplotlib implementation from scratch of machine learning algorithms used for classification.

The training set with N elements is defined as D={(X1, y1), . . ., (XN, yN)}, where X is a vector and y={0, 1} is one-hot encoded.

Sample code at the end of each file.

The Neural_Network_Derivatives.pdf document contains calculation of the derivatives used in code, except for the logistic regression that uses Yaoliang Yu lecture notes - see reference.

Contents

autoencoder: Autoencoder with sigmoid activation function in 2nd and 4th layers.

cnn: Flexible architecture of Convolutional Neural Network, with sigmoid and relu activation functions. Setup the number of layers:

• Convolution layer.
• Pooling layer.
• Full connected layer.

Note: code translated from Matlab to Python. Original code in https://github.com/ClodoaldoLima/Convolutional-Neural-Networks---Matlab

ensemble: Implementation of three ensemble methods of neural networks:

1. Ensemble learing via negative correlation.
2. Ensemble Learning Using Decorrelated Neural Networks.
3. Creating Diversity In Ensembles Using Artiflcial Data (DECORATE).

mixture of experts: Two setups of mixture of experts for time series:

1. Linear models for experts and gating with softmax output.
2. Linear models for experts and gating with normalized gaussian output.

neuralnets: Single Layer Perceptron (SLP) and Multi Layer Perceptron (MLP).

optimization: first and second order methods used in machine learning backpropagation. Methods available:

1. Gradient Descent.
2. Bisection.
3. Newton.
4. Modified Newton.
5. Levenberg-Marquardt.
6. Quasi-Newton Methods.
7. One Step Secant.
8. Conjugate Gradient Methods.

regression: implementation of three setups of regression for classification:

1. Linear regression.
2. Linear regression with regularization.
3. Logistic regression.

svm: implementation of three models of Support Vector Machines for binary and multi-class classification.

1. Traditional Support Vector Machines (SVM).
2. Least Squares Support Vector Machines (LSSVM).
3. Twin Support Vector Machines (TWSVM).

Kernel types:

1. Linear
2. Polynomial
3. Radial Base Function (RBF)
4. Exponential Radial Base Function (ERBF)
5. Hyperbolic tangent (tanh)
6. Linear splines

Acknowledgment

I would like to acknowledge professor Clodoaldo A. Moraes Lima for his guidance and support during the machine learning course at Universidade de Sao Paulo.