main.py

from __future__ import print_function
from __future__ import division
import json
import model_util
import numpy as np
import preprocess
import keras
from sklearn.model_selection import train_test_split
np.random.seed(123)  # for reproducibility

# set parameters:

subset = None

# Whether to save model parameters
save = True
#save=False
model_name_path = 'params/crepe_model3.h5'
model_weights_path = 'params/crepe_model_weights_with_test_v1.0.h5'

# Maximum length. Longer gets chopped. Shorter gets padded.
maxlen = 21

# Model params
# Filters for conv layers
nb_filter = 32
# Number of units in the dense layer
dense_outputs = 256
# Conv layer kernel size
filter_kernels = [3, 3, 2, 2, 2, 2]
# Number of units in the final output layer. Number of classes.
cat_output = 3

# Compile/fit params
batch_size = 750
nb_epoch = 1000

print('Loading data...')
# Expect x to be a list of sentences. Y to be index of the categories.
(xt, yt) = preprocess.load_data()

print('Creating vocab...')
vocab, reverse_vocab, vocab_size, alphabet = preprocess.create_vocab_set()

print('Build model...')
model = model_util.create_model(filter_kernels, dense_outputs, maxlen, vocab_size,
                              nb_filter, cat_output)
#model.add(Dropout(0.4))
#model=keras.models.load_model('params\crepe_model.h5','r+')
# Encode data
xt = preprocess.encode_data(xt, maxlen, vocab)
#X_train, X_test, Y_train, Y_test = train_test_split(xt, yt, test_size=0.3)
# x_test = preprocess.encode_data(x_test, maxlen, vocab)

print('Chars vocab: {}'.format(alphabet))
print('Chars vocab size: {}'.format(vocab_size))
print('X_train.shape: {}'.format(xt.shape))
#model.summary()
print('Fit model...')
model.fit(xt,yt,batch_size=batch_size, epochs=nb_epoch, shuffle=True)

prediction = model.predict(preprocess.encode_data(["laptop","obnoxious", "camembert", "portuguese", "penicillin", "protactinium"], maxlen, vocab))
print(prediction)

scores = model.evaluate(xt, yt)
print("\n%s: %.2f%%" % (model.metrics_names[1], scores[1]*100))
#import pickle
if save:
    print('Saving model params...')
    model.save_weights(model_weights_path)