multilabel, load_previous = sys.argv[1:]
print multilabel, load_previous
if multilabel == 'multi':
multilabel = True
else:
multilabel = False
if load_previous == 'load':
load_previous = True
else:
load_previous = False
train_set = Corpus(DATA_DIR + TRAIN_FILE, DATA_DIR + TRAIN_LABS)
X_train, X_val, y_train, y_val, nb_classes, word_index = load_data(
train_set, multilabel)
if load_previous:
model = load_model(STAMP, multilabel)
else:
model = build_model(nb_classes, word_index, EMBEDDING_DIM,
MAX_SEQUENCE_LENGTH, STAMP, multilabel)
if multilabel:
monitor_metric = 'val_f1_score'
else:
monitor_metric = 'val_loss'
return model
if __name__ == '__main__':
epochs = 2000
max_len = 40
batch_size = 512
data_sample = 0.5
skip = 3
STAMP = 'language_model'
data_set = Corpus(train_data,val_data,
max_len=max_len,
batch_size=batch_size,
data_sample=data_sample,
skip=skip)
with open('char2id.json', 'w') as fp:
json.dump(data_set.char2id, fp)
with open('id2char.json', 'w') as fp:
json.dump(data_set.id2char, fp)
model = build_model(max_len,data_set.vocab_size)
print(STAMP)
model_json = model.to_json()
with open(STAMP + ".json", "w") as json_file:
json_file.write(model_json)