Skip to content

Latest commit

 

History

History
14 lines (8 loc) · 700 Bytes

File metadata and controls

14 lines (8 loc) · 700 Bytes

Simple neural network playground I made to learn forward and backward propagation

The network consists of 2 hidden layers, each fallowed by an activation function, and outputs softmax probabilities for 2 classes. The training data type, training parameters and number of perceptron per layer can be changed.

The training progress of the network can be seen if the plot_output_during_training flag is set. Check out the jupyter notebook for more info.

Training data types: Linear, Circular, Check board

data_types

Sample NN output during training (tanh activation, circular training data):

nn-out