def predict_window(text_predict, number_predict, window_size):
"""
Generate a secuence to continue text_predict of len number_predict
param:
text_predict .- Init text
number_predict .- Number chars to generate
window_size .- The same that used in trainig
return
text_predict + number_predict chars
"""
# Get values used in trainig
chars_to_indices, indices_to_chars = load_coded_dictionaries()
number_chars = len(chars_to_indices)
# Clean input
input_clean = clean_text(text_predict.lower())
# Get the stub
channel = grpc.insecure_channel(IP + str(9000))
stub = prediction_service_pb2_grpc.PredictionServiceStub(channel)
# Call the service n times
for i in range(number_predict):
d = predict_one(
input_clean[i:], stub, window_size, number_chars,
chars_to_indices, indices_to_chars
)
input_clean += d
return input_clean
def test_load(self):
chars_to_indices_new, indices_to_chars_new = load_coded_dictionaries()
text_clean = load_text_clean('../data/Beiras.txt',
chars_to_indices_new)
text_to_csv(text_clean, "train.unittest", "test.unittest",
chars_to_indices_new, PERCENT_TRAIN)
text_read = csv_to_text("train.unittest", indices_to_chars_new)
self.assertEqual(text_clean[:101], text_read[0])
def predict(sentence, number_predict, window_size):
"""
Return a text sequence predicted by the GRU network continuing
the input sentence
:param
sentence: Input sentence
:return:
text sequence
"""
chars_to_indices, indices_to_chars = load_coded_dictionaries()
return predict_window(sentence, number_predict, window_size)
def predict(sentence):
"""
Return a text sequence predicted by the GRU network continuing
the input sentence
:param
sentence: Input sentence
:return:
text sequence
"""
chars_to_indices, indices_to_chars = load_coded_dictionaries()
model = create_gru_model(chars_to_indices)
model.load_weights(
'../model_weights/best_beiras_gru_textdata_weights.hdf5')
return predict_next_chars(model, sentence, window_size, chars_to_indices,
indices_to_chars)
def predict_window(text_predict, number_predict, window_size, lproject, lmodel,
version):
# Get dictionaries
chars_to_indices, indices_to_chars = load_coded_dictionaries()
# Get stub
service = googleapiclient.discovery.build('ml', 'v1')
name = 'projects/{}/models/{}'.format(lproject, lmodel)
if lversion is not None:
name += '/versions/{}'.format(lversion)
print(name)
input_clean = text_predict
# Call server for all charazters
for i in range(number_predict):
d = predict_one(input_clean[i:], service, name, window_size,
chars_to_indices, indices_to_chars)
input_clean += d
return input_clean
class TestPreproccess(unittest.TestCase):
def test_load(self):
chars_to_indices_new, indices_to_chars_new = load_coded_dictionaries()
text_clean = load_text_clean('../data/Beiras.txt',
chars_to_indices_new)
text_to_csv(text_clean, "train.unittest", "test.unittest",
chars_to_indices_new, PERCENT_TRAIN)
text_read = csv_to_text("train.unittest", indices_to_chars_new)
self.assertEqual(text_clean[:101], text_read[0])
if __name__ == "__main__":
"""
Generate FILE_OUTPUT_TRAIN and FILE_OUTPUT_TEST from '../data/Beiras.txt'
"""
# Load dictionaries to convert char to index and index to char
chars_to_indices_new, indices_to_chars_new = load_coded_dictionaries()
# Get String with clean text
text_clean = load_text_clean('../data/Beiras.txt', chars_to_indices_new)
# Generate files
text_to_csv(text_clean, FILE_OUTPUT_TRAIN, FILE_OUTPUT_TEST,
chars_to_indices_new, PERCENT_TRAIN)
# For test, read the file and test
text_read = csv_to_text(FILE_OUTPUT_TRAIN, indices_to_chars_new)
for sentence in text_read:
print(sentence)