ClassifyBaseModel.py - Used as a base class for different kind of neural networks
FullyConnectedNet.py - Fully Connected n layer neural network can be constructed using this
train_features of shape (N,[a,b,c])
train_labels os shape (N,)
validation_features - same shape as train features
validation_labels - same shape as train labels
CONFIG - contains the hyper parameter setup
- batch_size
- learning_rate
- learn_type - either of vanilla, adam, adagrad or rmsprop
- num_epochs - Number of epochs to train
- num_hidden_units - NUmber of hidden units in the hidden layers
- num_layers - NUmber of layers in the network
- log_folder - Folder into which the logs need to be written
- reg - regularization strength
- test_log_folder
- train_log_folder These are for storing logs that needs to be separated into test and train. For eg. Training versus validation accuracies