def attempt_init():
global init_status, init_error
try:
init_status = 'In Progress...'
from model.model import init
init()
init_status = 'Successful'
init_error = ''
except:
init_status = 'failed'
init_error = traceback.format_exc()
def main():
reload(sys)
sys.setdefaultencoding('utf-8')
init()
credit()
from model import model
from messages import input_message, output_message
if __name__ == "__main__":
# Parameters
cache_enabled = False
model_debug = False
steps = 1
step_output = True
# Initialize model
model = model("out/libmodel.so",
db_path="db_test",
cache_enabled=cache_enabled)
model.init()
# Setup model input
input_data = input_message()
print("Input:")
print(input_data)
# Prepare model output
output_data = output_message()
# Measure time
start_usec = datetime.datetime.now()
# Call model
for step in range(steps):
def initialize_model():
"""
Before running any tests, we will initialize the model.
"""
init()
def trainModel(params):
if params.features is None:
assert params.vocab, "No vocab provided"
assert params.trainData, "No trainData"
assert params.validationData, "No validationData"
assert params.modelFile, "No modelFile provided"
if params.features is None:
print("Vocab size :", len(params.vocab))
v_size = len(params.vocab)
else:
print("Features count :", params.features.len())
v_size = params.features.len()
print("Hidden layer :", params.hidden)
print("Word vec size :", params.wordVecSize)
print("Use GPU :", params.gpu)
print("Minibatches :", params.minibatches)
print("Models out dir :", params.modelFile)
print("Train Data :", params.trainData.len)
print("Validation Data:", params.validationData.len)
m = model.init(vocabularySize=v_size,
punctuationSize=len(data.PUNCTUATION_VOCABULARY),
hidden=params.hidden,
word_vector_size=params.wordVecSize,
optimizer=params.optimizer,
gpu=params.gpu,
use_features=params.features is not None)
m.summary(150)
# keras.utils.plot_model(m, 'punc.png')
# keras.utils.plot_model(m, 'punc_full.png', show_shapes=True)
if params.features is None:
gen_train = data.Generator(X=params.trainData.X,
y=params.trainData.y,
batch_size=params.minibatches)
gen_valid = data.Generator(X=params.validationData.X,
y=params.validationData.y,
batch_size=params.minibatches)
else:
gen_train = data.FeaturesGenerator(m_data=params.trainData,
features=params.features,
batch_size=params.minibatches)
gen_valid = data.FeaturesGenerator(m_data=params.validationData,
features=params.features,
batch_size=params.minibatches)
print("Training", file=sys.stderr)
checkpoint = ModelCheckpoint(filepath=params.modelFile,
monitor='loss',
verbose=1,
save_best_only=False,
mode='min',
period=1)
es = EarlyStopping(monitor='val_loss', mode='min', verbose=1)
callbacks = [checkpoint, es]
if params.callback is not None:
callbacks.insert(0, params.callback)
return m.fit_generator(
generator=gen_train,
validation_data=gen_valid,
epochs=params.maxEpochs,
verbose=1,
# workers=8,
# use_multiprocessing=True,
callbacks=callbacks)