test_model.py

import numpy as np
import sys
from keras.models import model_from_json
from keras.preprocessing import sequence
from keras.utils import np_utils
from keras.models import Sequential
from keras.layers.core import Dense, Dropout, Activation, TimeDistributedDense, RepeatVector
from keras.layers.embeddings import Embedding
from keras.layers import recurrent
from read_snli_data import *
from constants import *
from neural_net_helpers import *
from model import ModelDefinition

def main(model_name, test_filename, save_filename):
	# Read pickled files
	test_tuples = read_pickle_file(test_filename)
	X, Y = zip(*test_tuples)

	print 'Loading JSON file...'
	json_file = 'gru.json'
	model = model_from_json(open(json_file).read())
        print model
	print 'Loading weights...'
	model.load_weights(model_name + '.h5')
        model.compile(loss='categorical_crossentropy', optimizer='adam')

	print 'Loading wordsToInd...'
	f = open('wordsToInd.p', 'rb')
	wordsToIndices = pickle.load(f)
	f.close()

	print 'Loading Y_inds_to_words...'
	g = open('Y_inds_to_words.p', 'rb')
	Y_inds_to_words = pickle.load(g)
	g.close()
	max_length = model.layers[0].input_length

	print("Converting input sequences to lists of indices...")
	# Convert the input sentences to lists of lists of indices
	X_feats = convertToIndexSequence(X, wordsToIndices)
	X_feats = sequence.pad_sequences(X_feats, padding='pre', maxlen=max_length)
        
        a = open(save_filename, 'wb')
	for i, feat in enumerate(X_feats):
		print 'Example ' + str(i)
		example = np.array([X_feats[i, :]])
		preds = model.predict_classes(example)
		line = ' '.join(convert_to_word_list(preds, Y_inds_to_words)[0].split()[1:]) + '\n'
                a.write(line)
        a.close()


if __name__ == '__main__':
	main(sys.argv[1], sys.argv[2], sys.argv[3])