This repository contains python codes of machine learning modules

FOLDER: NeuralNetwork_implementation/

neural_network.py

class "classification"

Feedforward artificial network. Using gradient descent, forward, backward propagation to optimize weights.
Weights between neurons are initialized randomly in the range of [-4/sqrt(d), 4/sqrt(d))
See example_neural_network.py for an example of using the code

Input Parameters

• nf: number of features
• hidden_sl: a tuple with # of neurons of layers
• lrate: learning rate (default: 60.)
• penalty: l2 regularization penalty parameter (default: 0.0001)
• maxiter: maximum # of iterations (default: 50000)
• tol: tolerance value of cost function (default: 1e-6)

Attributes

• fit(x, y): do the neural network fitting
• predict(): input a 1-D array of x, output predicted y value with probability

Example codes

example_neural_network.py

An example code for using the neural_network.classfication() class by inputing a set of XOR data:
0 0 False
0 1 True
1 0 True
1 1 False
The fitting results (probability map) can be seen in
fittiingResult_exNeuralNetwork1_1.png and fittingResult_exNeuralNetwork1_2.png

example_logistic.py

An example code for using scikit-learn to do logistic fitting, with input of data.xyz (my simulation data)
The fitting results can be seen in fitting_result_exLogistic.png, where (yellow, purple) represent (True, False), and 3 ovals are 0.2, 0.5, 0.8 from the fitting results of logistic regression

FOLDER: EmailClassification_sklearn_naiveBayes/

This folder contains an email classification code:
Using naive Bayes method to determine if an email belongs to author 1 or author 2