ML

ML from Scratch

Polynomial Regression

Description (main.py)

• This code does not use sklearn functions (from scratch)
• This code generate a polynomial (degree k) feature map (0 to k) using the training data (function create_poly).
• This code generate a polynomial (degree k) feature map (0 to k) plus sinusoidal function using the training data (function create_sin).
• This code fit linear model solving Ax = b (function fit)
• This code make a prediction given new inputs (function predict).

Execution

python main.py

Description (vscikit-learn.py)

• This version of code uses sklearn functions to valid the previous results.

Execution

python vscikit-learn.py

Logistic Regression

Description (main.py)

• This code does not use sklearn functions (from scratch)
• This code fits a model using Newton's method to minimize the logistic regression loss function.
• This code make a prediction given new inputs (function predict).

Execution

python main.py

Description (vscikit-learn.py)

• This version of code uses sklearn functions to valid the previous results.

Execution

python vscikit-learn.py

Gaussian Discriminant Analysis

Description (main.py)

• This code does not use sklearn functions (from scratch)
• Parameters associated to Multivariate Gaussian distribution are computed: mean vectors (mu) and covariance matrix (sigma).
• This code make a prediction given new inputs (function predict).

Execution

python main.py

Description (vscikit-learn.py)

• This version of code uses sklearn functions to valid the previous results.

Execution

python vscikit-learn.py

Poisson Regression

Description (main.py)

• This code does not use sklearn functions (from scratch)
• This code implements a Generalized Linear Model with a Poisson distribution with gradient ascent.
• This code make a prediction given new inputs (function predict).

Execution

python main.py

Description (vscikit-learn.py)

• This version of code uses sklearn functions to valid the previous results.

Execution

python vscikit-learn.py

Naive Bayes Classifier

Description (main.py)

• This code does not use sklearn functions (from scratch)
• This code classifies (predicts) a message text is spam or not.
• This code make a prediction given new inputs (function predict_from_naive_bayes_model).

Execution

python main.py

Description (vscikit-learn.py)

• This version of code uses sklearn functions to valid the previous results.

Execution

python vscikit-learn.py

Support Vector Machines

Description (main.py)

• This code does not use sklearn functions (from scratch)
• This code classifies (predicts) a message text is spam (1) or not (0).
• This code make a prediction given new inputs (function svm.train_and_predict_svm).

Execution

python main.py

Description (vscikit-learn.py)

• This version of code uses sklearn functions to valid the previous results.

Execution

python vscikit-learn.py

K-means

Description (main.py)

• This code does not use sklearn functions (from scratch)
• This code uses k-Means to compress a image. Where the image is reduced to 16 colors (num_clusters = 16).
• Each pixel (RGB) is assigned to the closest centroid.

Execution

python main.py

Description (vscikit-learn.py)

• This version of code uses sklearn functions to valid the previous results.

Execution

python vscikit-learn.py

Semi Supervised Expectation Maximization

Description (main.py)

• This code does not use sklearn functions (from scratch)
• This code implements Gaussian Mixture Model (GMM) to apply semi-supervised EM algorithm with labelled and unlabelled data.
• E-step and M-step are computed.
• The input data (or new data) is clustered.

Execution

python main.py

Description (vscikit-learn.py)

• This version of code uses sklearn functions to valid the previous results (Unsupervised Expectation-Maximization).

Execution

python vscikit-learn.py

Neural Network

Description (main.py)

• This code does not use sklearn functions (from scratch)
• This code implements a simple neural network (NN) to classify grayscale images of handwritten digits (0 - 9) from the MNIST dataset.
• This NN has a single hidden layer (sigmoid function as activation) and cross entropy loss (softmax function for the output layer).

Execution

python main.py

Description (vTensorFlow.py)

• This version of code uses TensorFlow functions to valid the previous results.

Execution

python vTensorFlow.py

Reinforcement Learning - The inverted pendulum

Description (cartpole.py)

• This code does not use sklearn functions (from scratch)
• Parts of the code (cart and pole dynamics, and the state discretization) are inspired from code available at the RL repository http://all.cs.umass.edu//rlr//domains.html and the course XCS229 Machine Learning from Stanford University.
• The cart-pole system is described in cartpole.py. The main simulation loop in this file calls the simulate() function for simulating the pole dynamics, get_state() for discretizing the otherwise continuous state space in discrete states, and show_cart() for display.
• The code presented in functions.py shows a estimate model for the underlying Markov Decision Process (MDP), solving the Bellman's equations for this estimated MDP.

Execution

python cartpole.py

Reinforcement Learning - Grid World environment

Description

• This code does not use sklearn functions (from scratch)
• This code implements REINFORCE policy gradient algorithm to optimize on a stochastic policy that solves a simple gridded environment.

Execution

python train.py