parser.add_argument('--X_train_train_npy', default='data/X_train_train')
parser.add_argument('--X_train_test_npy', default='data/X_train_test')
parser.add_argument('--y_train_train_npy', default='data/y_train_train')
parser.add_argument('--y_train_test_npy', default='data/y_train_test')
parser.add_argument('--hw', default=48, type=int)
parser.add_argument('--out_dir', default='models/')
args = parser.parse_args()
if __name__ == '__main__':
print('Loading training images')
X_train, X_test = np.load('%s_%i.npy' % (args.X_train_train_npy, args.hw)), np.load('%s_%i.npy' % (args.X_train_test_npy, args.hw))
y_train, y_test = np.load('%s_%i.npy' % (args.y_train_train_npy, args.hw)), np.load('%s_%i.npy' % (args.y_train_test_npy, args.hw))
print(X_train.shape, X_test.shape, y_train.shape, y_test.shape)
print('Loading base net from %s' % args.base_net_fname)
base_net = utils.load_from_pickle(args.base_net_fname)
print('Loading model definition from %s' % args.net_name)
net = get_net(args.net_name)
net.load_weights_from(base_net)
t0 = time()
print('Started training at %s' % t0)
net.fit(X_train, y_train)
print('Finished training. Took %i seconds' % (time() - t0))
y_test_pred = net.predict(X_test)
y_test_pred_proba = net.predict_proba(X_test)
lscore = utils.multiclass_log_loss(y_test, y_test_pred_proba)
ascore = net.score(X_test, y_test)
parser.add_argument('--translate_y_upper', default=-2, type=int)
parser.add_argument('--translate_y_step', default=3, type=int)
parser.add_argument('--translate_x_lower', default=2, type=int)
parser.add_argument('--translate_x_upper', default=-2, type=int)
parser.add_argument('--translate_x_step', default=3, type=int)
args = parser.parse_args()
if __name__ == '__main__':
print('Loading test images from %s' % (args.X_train_npy))
X = np.load(args.X_train_npy)
hw = args.hw
print('Loading model from %s' % args.model)
net = utils.load_from_pickle(args.model)
# if 'mean' not in net.batch_iterator_test:
# print('Warning: net.batch_iterator_test does not have preset mean value. Using mean from all training data for now')
net.batch_iterator_test.mean = np.mean(X, axis=0)
scale_choices = np.linspace(args.scale_lower, args.scale_upper,
args.scale_step)
rotation_choices = range(args.rotation_lower, args.rotation_upper,
args.rotation_step)
translate_y_choices = range(args.translate_y_lower, args.translate_y_upper,
args.translate_y_step)
translate_x_choices = range(args.translate_x_lower, args.translate_x_upper,
args.translate_x_step)
combinations = list(
itertools.product(scale_choices, rotation_choices, translate_y_choices,