return loss, metric
#############################
# main function boilerplate #
#############################
if __name__=='__main__':
# create model
model = create_convnet_cifar10_model(num_classes=10)
# declare the model's input dimension
# Training does not require this, but it is needed for deployment.
model.update_signature((num_channels, image_height, image_width))
# criterion function. This is what is being trained trained.
# Model gets "sandwiched" between normalization (not part of model proper) and criterion.
criterion = create_criterion_function(model, normalize=lambda x: x / 256)
# train
reader = create_reader(os.path.join(data_path, 'train_map.txt'), os.path.join(data_path, 'CIFAR-10_mean.xml'), True)
train_model(reader, model, criterion, max_epochs=80)
# save and load (as an illustration)
path = data_path + "/model.cmf"
model.save(path)
# test
model = Function.load(path)
reader = create_reader(os.path.join(data_path, 'test_map.txt'), os.path.join(data_path, 'CIFAR-10_mean.xml'), False)
criterion = create_criterion_function(model, normalize=lambda x: x / 256)
evaluate(reader, criterion)
def criterion(x, y):
z = model(normalize(x))
ce = cross_entropy_with_softmax(z, y)
errs = classification_error(z, y)
return (Function.NamedOutput(loss=ce),
Function.NamedOutput(metric=errs))