simpleMLP

A very simple Python code to generate and train 1 or 2-layer fully connected neural network (MLP)

The MLP relies only on Python and Numpy for calculation and matplotlib for display of graphs

How to use

Everything must be done in the test.py file. neuralnetwork.py includes the algorithms necessary for pass-forwarding, retro-propagation and formatting tools.

Loading your data

All data must be in the form of numpy.array. Sample data must be offloat type, while Labels must be of int type.

• x = training data. Shape = [dimensions of sample, examples]
• labels = training labels. Shape = [nb of classes, examples]
• xtest = test data. Same shape as x.
• labeltest = test labels. Same shape as labels

Parameters

• c = number of hidden-layer perceptrons (only used in 2-layers MLP)
• lr = learning rate
• it_train = number of training iterations between each test on the test database
• epoch = number of test iterations

Therefore, the total number of training is it_train * epoch.

Choose between 1 layer or 2 layers

In test.py, 2 successive zones of code represents the training for 1-layer and 2-layers MLP. Simply comment the one you don't want to use.