def prepare_trainer(net,
optimizer_name,
momentum,
lr,
num_gpus,
state_file_path=None):
optimizer_name = optimizer_name.lower()
if (optimizer_name == "sgd") or (optimizer_name == "nag"):
optimizer = keras.optimizers.SGD(lr=lr,
momentum=momentum,
nesterov=(optimizer_name == "nag"))
else:
raise ValueError("Usupported optimizer: {}".format(optimizer_name))
backend_agnostic_compile(model=net,
loss="categorical_crossentropy",
optimizer=optimizer,
metrics=[
keras.metrics.categorical_accuracy,
keras.metrics.top_k_categorical_accuracy
],
num_gpus=num_gpus)
if (state_file_path is not None
) and state_file_path and os.path.exists(state_file_path):
net = load_model(filepath=state_file_path)
return net
def test(net,
val_gen,
val_size,
batch_size,
num_gpus,
calc_weight_count=False,
extended_log=False):
"""
Main test routine.
Parameters:
----------
net : Model
Model.
val_gen : generator
Data loader.
val_size : int
Size of validation subset.
batch_size : int
Batch size.
num_gpus : int
Number of used GPUs.
calc_weight_count : bool, default False
Whether to calculate count of weights.
extended_log : bool, default False
Whether to log more precise accuracy values.
"""
keras.backend.set_learning_phase(0)
backend_agnostic_compile(
model=net,
loss="categorical_crossentropy",
optimizer=keras.optimizers.SGD(
lr=0.01,
momentum=0.0,
decay=0.0,
nesterov=False),
metrics=[keras.metrics.categorical_accuracy, keras.metrics.top_k_categorical_accuracy],
num_gpus=num_gpus)
# net.summary()
tic = time.time()
score = net.evaluate_generator(
generator=val_gen,
steps=(val_size // batch_size),
verbose=True)
err_top1_val = 1.0 - score[1]
err_top5_val = 1.0 - score[2]
if calc_weight_count:
weight_count = keras.utils.layer_utils.count_params(net.trainable_weights)
logging.info("Model: {} trainable parameters".format(weight_count))
if extended_log:
logging.info("Test: err-top1={top1:.4f} ({top1})\terr-top5={top5:.4f} ({top5})".format(
top1=err_top1_val, top5=err_top5_val))
else:
logging.info("Test: err-top1={top1:.4f}\terr-top5={top5:.4f}".format(
top1=err_top1_val, top5=err_top5_val))
logging.info("Time cost: {:.4f} sec".format(
time.time() - tic))
def test(net,
val_gen,
val_size,
batch_size,
num_gpus,
calc_weight_count=False,
extended_log=False):
keras.backend.set_learning_phase(0)
backend_agnostic_compile(model=net,
loss='categorical_crossentropy',
optimizer=keras.optimizers.SGD(lr=0.01,
momentum=0.0,
decay=0.0,
nesterov=False),
metrics=[
keras.metrics.categorical_accuracy,
keras.metrics.top_k_categorical_accuracy
],
num_gpus=num_gpus)
# net.summary()
tic = time.time()
score = net.evaluate_generator(generator=val_gen,
steps=(val_size // batch_size),
verbose=True)
err_top1_val = 1.0 - score[1]
err_top5_val = 1.0 - score[2]
if calc_weight_count:
weight_count = keras.utils.layer_utils.count_params(
net.trainable_weights)
logging.info('Model: {} trainable parameters'.format(weight_count))
if extended_log:
logging.info(
'Test: err-top1={top1:.4f} ({top1})\terr-top5={top5:.4f} ({top5})'.
format(top1=err_top1_val, top5=err_top5_val))
else:
logging.info('Test: err-top1={top1:.4f}\terr-top5={top5:.4f}'.format(
top1=err_top1_val, top5=err_top5_val))
logging.info('Time cost: {:.4f} sec'.format(time.time() - tic))
