def findPreviouslyTrainedModel(model_name):
model_path = models_path(model_name + zipped_pickle_extension)
if os.path.isfile(model_path):
return model_name, 0
epoch_string = "_e_"
base_name, number_of_epochs = model_name.split(epoch_string)
number_of_epochs = int(number_of_epochs)
previous_model_names = [
m.replace(zipped_pickle_extension, "")
for m in os.listdir(models_path()) if base_name in m
]
previous_numbers_of_epochs = [
int(m.split(epoch_string)[-1]) for m in previous_model_names
]
previous_numbers_of_epochs = [
e for e in previous_numbers_of_epochs if e <= number_of_epochs
]
previous_model_names = previous_model_names[:len(previous_numbers_of_epochs
)]
if len(previous_model_names) == 0:
return None, number_of_epochs
previous_model_epochs_trained, previous_model_name = \
sorted(zip(previous_numbers_of_epochs, previous_model_names))[-1]
epochs_still_to_train = number_of_epochs - previous_model_epochs_trained
return previous_model_name, epochs_still_to_train
def saveModel(model, model_name):
model_path = models_path(model_name + zipped_pickle_extension)
print("Saving model parameters and metadata.")
with gzip.open(model_path, "wb") as model_file:
pickle.dump(model, model_file)
print("Model parameters and metadata saved in {}.".format(model_path))
def loadModel(model_name):
epoch = model_name.split("_e_")[-1]
model_path = models_path(model_name + zipped_pickle_extension)
print("Loading model parameters for epoch {} and metadata.".format(epoch))
with gzip.open(model_path, "rb") as model_file:
model = pickle.load(model_file)
print("Model parameters and metadata loaded.")
return model
def modelTrained(model_name):
model_path = models_path(model_name + zipped_pickle_extension)
return os.path.isfile(model_path)