Multilayer Neural Network in Python

Python implementation of multilayer perceptron neural network from scratch.

Minimal neural network class with regularization using scipy minimize. Contains clear pydoc for learners to better understand each stage in the neural network.

Requirements

• Python 3.4 (tested)

Installation

    pip install mlperceptron

Goal

To provide an example of a simple MLP for educational purpose.

Code sample

Predicting outcome of AND logic gate:

X = 000, 001, 010, 011, 100, 101, 110, 111 y = 0,0,0,0,0,0,1

Data we want to predict: p = 011, 111, 000, 010, 111 Expected results are: 0, 1, 0, 0, 1

    import numpy as np
    from mlperceptron.mlperceptron import NeuralNetwork

    X = np.matrix(
        '0 0 0;0 0 1;0 1 0;0 1 1;1 0 0;1 0 1;1 1 0;1 1 1')
    y = np.matrix('0;0;0;0;0;0;0;1')
    n = NeuralNetwork((5,5,))

    g = n.train(X, y, 0.01, show_cost=True)
    y_pred = n.predict(np.matrix('0 1 1;1 1 1;0 0 0;0 1 0;1 1 1'), g)

    print(y_pred)
    print(n.accuracy(y_pred, np.matrix('0;1;0;0;1')))

Result

Loss: 1.150514159559136
Loss: 1.1505141504773504
Loss: 1.1505141583532519
Loss: 1.1505141469667088
Loss: 1.2434056388432244
Loss: 0.8288594235909958
Loss: 0.8288594198770226
Loss: 0.8288594228018156
Loss: 0.7807394496223645
....
....
Loss: 0.11768930150540982
Loss: 0.11768930150555365
Loss: 0.11768930150553429
Loss: 0.1176893015053323
Loss: 0.11768328608538177
Loss: 0.1176832860852971
Loss: 0.11768328608527757
Loss: 0.11768328608535958
Loss: 0.1176832860853596
Loss: 0.11768328608534921
Loss: 0.1176832860853433
Loss: 0.11768197870406577
Loss: 0.11768197870406577
[[0]
 [1]
 [0]
 [0]
 [1]]
1.0

Contributors

• Paulo Kuong (@pkuong)