mllb 0.0.3

Machine Learning package

mllb

Machine Learning algorithms.

pip install mllb

Overview

This repository contains Machine Learning algorithms for Linear Regression, Logistic Regression, and Convolutional Neural Networks. It also helps to easily and efficiently port, edit and select rows in csv files directly for training.

This project was inpired by the author's quest to understand the basic concepts and mathematics of Machine Learning.

It uses NumPy heavily for its computations.

Basics

Here, the basic a few basic things that can be done with the package are discussed. A more detailed documentation can be found in the docs directory.

Linear Regression

To train a simple Linear Regression algorithm from x, and y.

import mllb as ml

x = [1, 2, 3, 4, 5, 6]
y = [3, 5, 7, 9, 11, 13]

# Create a linear regression model
model = ml.linear_regression(x, y)

# Choose a learning rate
model.learning_rate.initial(0.1)

# Choose the number of epochs and train, the default optimizer is Stochastic Gradient Descent
model.train(10)

# Test the model by predicting a value
pr = model.predict([10])

# Print the prediction
print(pr)

Convolutional Neural Network

To train a simple Convolutional Neural Neural network, four simple 'pictures' whose resolutions are 4 x 4 pixel pictures will be used as an example here.

image 1 = [white, white]
          [black, black]

image 2 = [black, white]
          [white, black]

image 3 = [black, black]
          [white, white]

image 4 = [white, black]
          [black, white]

import mllb as ml

# Using white = 1 and black = 0, the four images can be represented as x
x = [[1, 1, 0, 0], [0, 1, 1, 0], [0, 0, 1, 1], [1, 0, 0, 1]]

# Choose a corresponding value y that represents the classifier. Here, we choose:
# image 1 as [0, 0]; image 2 as [0, 1]; image 3 as [1, 0] and image 4 as [1, 1]
y = [[0, 0], [0, 1], [1, 0], [1, 1]]

# Create a cnn (Convolutional Neural Network model)
model = ml.cnn(x, y)

# Create the layers for the network using the activation function names
# Here two layers will be created; a ReLU layer with 5 nodes, and a sigmoid layer with 2 nodes
model.layers.create(ml.relu(5), ml.sigmoid(2))

# Intialize the layer weights with xavier method for faster convergence
model.layers.xavier_initialization()

# Set an intial learning rate
model.learning_rate.initial(0.01)

# If you wish, the learning rate can be multiplied by a value after every epoch for faster convergence.
# This part optional. Here we use 1.01 (increase learning rate by 1% after every epoch)
model.learning_rate.multiply_after_epoch(1.01)

# Train the model and choose the number of epochs, an epoch of 800 is chosen here. It defaults to Stochastic Gradient Descent.
model.train(800)

# Test your model by predicting an image
img1 = model.predict([1, 1, 0, 0])
img2 = model.predict([0, 1, 1, 0])
img3 = model.predict([0, 0, 1, 1])
img4 = model.predict([1, 0, 0, 1])

# Print the predictions
print(img1)
print(img2)
print(img3)
print(img4)

# Print the predictions
print(img1)
print(img2)
print(img3)
print(img4)