Multi-Layer-Perceptron

This repository has an elementary implementation of a Multi Layer perceptron with 3 Layers and 5 Nodes in the hidden layer

Note: Implementation has input with 785 features. Sorry ! couldn't accomodate that many in the picture.

The implementation is completely done using numpy.

Directory and files description:

1. data: This has the datasets used for training and testing the model.

2. acc_calc.py: utility for checking accuracy.

3. nn.py: This file has the training algorithms including backpropagation process

Utilities defined and their usages:

• calculateSigmoid ! calculates sigmoid of every element in the matrix.
• calculateMatrixDotProduct ! calculates dot product of two given matrices.
• calculateSigmoidDerivative ! Calculates (1 - sigMatrix)*sigMatrix, (1-O)*O
• predict ! predicts a data sample given input and relevant weights, returns in one hot encoded format.
• errorCumulative!calculates total error of all data samples given target(t) and output(O).
• calculateYofLayer ! calculates sigmoid(WtX) of a any given layer.
• calculateAccuracy ! calculates accuracy given target(t) and output(O).
• addBiasTerm ! adds bias term 1 to all the data samples passed as a matrix
• trainModel !

1. Loads the training data and labels.
2. Initializes random weights for the start.
3. Runs epochs with feed forward and backpropagation logic till error value is below a considered value.
4. Check accuracy on validate dataset.
5. Saves the trained weights.

test_mlp.py

Utilities defined and their usages:

• Takes path of directory where test data exists.
• Loads the model.
• calls tarinModel from nn.py and gets the weights saved
• Predicts the data test samples.

Sample console output

How do i run it ?

python test_mlp.py

Watch the legendary neural network in action ! haha.. cheers