logging.info("DONE")
np.save(TEST_DATASET_NPY_PATH, test_dataset)
logging.info("Saving test dataset at {}".format(
os.path.dirname(TEST_DATASET_NPY_PATH)))
else:
logging.info("Loading from serialized files at {}".format(
os.path.dirname(TRAIN_DATASET_NPY_PATH)))
test_dataset = np.load(TEST_DATASET_NPY_PATH)
return train_dataset, train_labels, test_dataset
if __name__ == "__main__":
logging.info("STARTING digit-recog model")
train_dataset, train_labels, test_dataset = get_datasets()
train_dataset = train_dataset.reshape((train_dataset.shape[0], 28, 28, 1))
train_labels = train_labels.reshape((train_dataset.shape[0], 10, 1))
# mlp = MLP([784, 120, 84, 10])
# mlp.SGD(train, 30, alpha=0.1, batch_size=10)
cnn_model = CNN([('conv', 5, 6, 2, 1, 'sigmoid'),
('pool', 'average', 2, 2),
('conv', 5, 16, 0, 1, 'sigmoid'),
('pool', 'average', 2, 2),
('fully-connected', 120, 'sigmoid'),
('fully-connected', 84, 'sigmoid'),
('fully-connected', 10, 'sigmoid')], (28, 28, 1))
cnn_model.SGD(train_dataset, train_labels, 30, 3.)
