def evaluation(csv_filename: str,
model_dir: str):
config_filename = os.path.join(model_dir, CONST.CONFIG_FILENAME)
parameters = Params.from_json_file(config_filename)
saving_dir = os.path.join(parameters.output_model_dir, CONST.FOLDER_SAVED_MODEL)
# Callback for model weights loading
last_time_stamp = max([int(p.split(os.path.sep)[-1].split('-')[0])
for p in glob(os.path.join(saving_dir, '*'))])
loading_dir = os.path.join(saving_dir, str(last_time_stamp))
ld_callback = CustomLoaderCallback(loading_dir)
# Pre-process csv data
csv_evaluation_file = os.path.join(parameters.output_model_dir,
CONST.PREPROCESSING_FOLDER, CONST.EVALUATION_FILENAME)
n_samples = preprocess_csv(csv_filename,
parameters,
csv_evaluation_file)
dataset_evaluation = dataset_generator([csv_evaluation_file],
parameters,
batch_size=parameters.eval_batch_size,
num_epochs=1)
# get model and evaluation
model = get_model_train(parameters)
eval_output = model.evaluate(dataset_evaluation,
callbacks=[ld_callback])
print('-- Metrics: ', eval_output)
def prediction(csv_filename: str, output_model_dir: str):
parameters = Params.from_json_file(os.path.join(output_model_dir, CONST.CONFIG_FILENAME))
saving_dir = os.path.join(output_model_dir, CONST.FOLDER_SAVED_MODEL)
last_time_stamp = str(max([int(p.split(os.path.sep)[-1].split('-')[0])
for p in glob(os.path.join(saving_dir, '*'))]))
model = get_model_inference(parameters, os.path.join(saving_dir, last_time_stamp, CONST.MODEL_WEIGHTS_FILENAME))
data_set_test = dataset_generator([csv_filename],
parameters,
use_labels=False,
batch_size=parameters.eval_batch_size,
shuffle=False)
ps_callback = CustomPredictionSaverCallback(output_model_dir, parameters)
_, _, _ = model.predict(x=data_set_test, callbacks=[ps_callback])
def training(_config: dict):
parameters = Params(**_config)
export_config_filename = os.path.join(parameters.output_model_dir,
CONST.CONFIG_FILENAME)
saving_dir = os.path.join(parameters.output_model_dir,
CONST.FOLDER_SAVED_MODEL)
is_dir_exist = os.path.isdir(parameters.output_model_dir)
is_dir_del = parameters.del_output_model_dir
is_dir_restore = parameters.restore_model
if not is_dir_exist:
is_dir_restore = False
os.makedirs(parameters.output_model_dir)
elif is_dir_del:
is_dir_restore = False
shutil.rmtree(parameters.output_model_dir)
os.makedirs(parameters.output_model_dir)
elif not is_dir_restore:
assert not is_dir_exist, \
'{} already exists, you cannot use it as output directory.'.format(parameters.output_model_dir)
os.makedirs(parameters.output_model_dir)
# data and csv pre-processing
csv_train_file, csv_eval_file, \
n_samples_train, n_samples_eval = data_preprocessing(parameters)
parameters.train_batch_size = min(parameters.train_batch_size,
n_samples_train)
parameters.eval_batch_size = min(parameters.eval_batch_size,
n_samples_eval)
# export config file in model output dir
with open(export_config_filename, 'w') as file:
json.dump(parameters.to_dict(), file)
# Create callbacks
log_dir = os.path.join(parameters.output_model_dir, 'logs')
os.makedirs(log_dir, exist_ok=True)
tb_callback = tf.keras.callbacks.TensorBoard(log_dir=log_dir,
profile_batch=0)
lrtb_callback = LRTensorBoard(log_dir=log_dir, profile_batch=0)
lr_callback = tf.keras.callbacks.ReduceLROnPlateau(factor=0.5,
patience=10,
cooldown=0,
min_lr=1e-8,
verbose=1)
es_callback = tf.keras.callbacks.EarlyStopping(min_delta=0.005,
patience=20,
verbose=1)
sv_callback = CustomSavingCallback(saving_dir,
saving_freq=parameters.save_interval,
save_best_only=True,
keep_max_models=3)
list_callbacks = [
tb_callback, lrtb_callback, lr_callback, es_callback, sv_callback
]
if is_dir_restore:
last_time_stamp = max([
int(p.split(os.path.sep)[-1].split('-')[0])
for p in glob(os.path.join(saving_dir, '*'))
])
loading_dir = os.path.join(saving_dir, str(last_time_stamp))
ld_callback = CustomLoaderCallback(loading_dir)
list_callbacks.append(ld_callback)
with open(os.path.join(loading_dir, CONST.EPOCH_FILENAME), 'rb') as f:
initial_epoch = pickle.load(f)
epochs = initial_epoch + parameters.n_epochs
else:
initial_epoch = 0
epochs = parameters.n_epochs
# Get model
model = get_model_train(parameters)
# Get data-sets
data_set_train = dataset_generator(
[csv_train_file],
parameters,
batch_size=parameters.train_batch_size,
data_augmentation=parameters.data_augmentation,
num_epochs=parameters.n_epochs)
data_set_eval = dataset_generator([csv_eval_file],
parameters,
batch_size=parameters.eval_batch_size,
data_augmentation=False,
num_epochs=parameters.n_epochs)
# Train model
model.fit(data_set_train,
epochs=epochs,
initial_epoch=initial_epoch,
steps_per_epoch=np.floor(n_samples_train /
parameters.train_batch_size),
validation_data=data_set_eval,
validation_steps=np.floor(n_samples_eval /
parameters.eval_batch_size),
callbacks=list_callbacks)