main.py

from Model import Model
import numpy as np
import pickle

# Note: Total sentences are 2113 in number.

if __name__ == '__main__':

    #####################################
    # don't change this
    np.random.seed(43)
    # hyper parameters to be tweaked here
    load_rnn_from_pickle = False
    training_size = 2113  # maximum of 2113
    l_rate = 0.01
    mini_batch_size = 2
    reg_cost = 0.001
    epochs = 100
    activation_func = "tanh"
    dim = 100
    ######################################

    # perform 60-15-25 percent split into train-val-test sets
    train = np.ceil(0.6 * training_size)
    val = np.ceil(0.15 * training_size)
    test = training_size - train - val

    # load parsed trees from file in PTB format
    if load_rnn_from_pickle is False:
        with open('WikiTreebankQuartiles_second.txt', 'rb') as fh:
            RNN = Model(dim=dim, l_rate=l_rate, mini_batch=mini_batch_size, reg_cost=reg_cost,
                        epochs=epochs, activation_func=activation_func)

            for i, line in enumerate(fh):
                p = line.find(' ')
                ptb_string = line[p + 1:]
                rid = line[:p]
                # Add to the list of trees
                RNN.add_tree(ptb_string, rid)

        with open('rnn.pickle', 'wb') as pickle_file:
            pickle.dump(RNN, pickle_file, pickle.HIGHEST_PROTOCOL)
    else:
        with open('rnn.pickle', 'rb') as pickle_file:
            RNN = pickle.load(pickle_file)

    indices = np.arange(0, training_size)
    # create separate indices for the 3 data sets
    np.random.shuffle(RNN.trees)
    np.random.shuffle(indices)
    RNN.tree_train = indices[:train]
    RNN.tree_val = indices[train:train + val]
    RNN.tree_test = indices[train + val:]
    # print RNN.cross_validate()
    RNN.train(True)
    # RNN.check_model_veracity()
    print "Test Cost Function, Accuracy, Incorrectly classified sentence Ids"
    print RNN.test()

    hyper_params = "training_size={0}\nl_rate={1}\nmini_batch_size={2}\nreg_cost={3}\nepochs={4}\ndim={5}\nactivation_func={6}".format(
        training_size, l_rate, mini_batch_size, reg_cost, epochs, dim, activation_func)
    print hyper_params