def multi_driver_training(options):
""" Train model using all drivers' data in the given datapath."""
print("Loading data...")
dataset = load_all(options)
print("Building model and compiling functions...")
net = models.cnn_1(options, output_size=len(dataset['label_map']))
# net = models.softmax_only(output_size=len(dataset['label_map']))
print("Starting training...")
start_time = time.time()
try:
train_loop(dataset['train_data'], dataset['val_data'], net, options)
print('Training Complete...')
except KeyboardInterrupt:
print('Keyboard Interrupt...')
end_time = time.time()
print('--------------------')
print(' Saving model, check logs for results.\n'
' Time taken: {0}\n'
.format(end_time - start_time))
utils.save_model(model=get_all_param_values(net), options=options)
return net
def per_driver_train_predict(options):
""" Predict output probabilities for each driver.
If `options.train` is true, this trains a model"""
# TODO: support for predict only
# TODO: support for pretrained weights
print("Loading data...")
dataset = load_all(options)
all_outputs = deque([])
for driver_index in range(len(dataset['label_map'])):
driver_id = dataset['label_map'][driver_index]
print('Working on Driver {0}'.format(driver_id))
net = _train_single(driver_index, dataset, options)
utils.save_model(model=get_all_param_values(net), options=options, driver_id=driver_id)
print('Training complete for driver {0}, making predictions...'.format(driver_id))
outputs = _predict_single(net, driver_id, options)
utils.save_outputs(outputs, options, driver_id=driver_id)
all_outputs.extend(outputs)
utils.save_outputs(all_outputs, options)
return