matrajaya / deep-learning-models

Implementing from scratch: NN, RNN, LSTM, GRU and CNN using theano.

Deep-Learning-Models

Implementing from scratch: NN, RNN, LSTM, GRU and CNN using theano.

Usage:

import rnn
model = rnn.rnn_theano(vocab_size = 2000,h_dim = 100 ,saved_model = False)

vocab_size = size of vocabulary If using a previously saved model, use saved_model=True

The format the data should be

data = ['crystals in urine results',
        'picture of state trooper motorcycles',
        'chester a arthur',
        'missouri dept of elementary and secondary',
        'business forms for year end statement']

Prepare the data using

X_train,Y_train = rnn.prepare_data(data)

Train the model

rnn.train_with_sgd(model,X_train,Y_train,nepoch=3,learning_rate=0.01)

For LSTM and GRU

import lstm,gru
model = lstm.lstm_theano(vocab_size = 2000,h_dim = 100)
model = gru.gru_theano(vocab_size = 2000,h_dim = 100)