train_input = np.vstack([train_input, v])
# if len(parts) == 5 and parts[2] in ['7', '8']:
# cv_input = np.vstack([cv_input, v])
# cv_output = np.vstack([cv_output, o])
# elif len(parts) == 5 and parts[2] in ['5', '6']:
# test_input = np.vstack([cv_input, v])
# test_output = np.vstack([cv_output, o])
# else:
# train_input = np.vstack([train_input, v])
# train_output = np.vstack([train_output, o])
rbm = RBM(841, 100,
rng=(lambda n: np.random.normal(0, 0.001, n)))
print(rbm)
rbm.train(train_input,
cd_k=1,
learning_rate=0.01,
momentum_rate=0.9,
max_iter=1000,
batch_size=10,
n_iter_stop_skip=10,
goal=hamming,
#cv_input_data=cv_input,
stop_threshold=0.15,
#neural_local_gain=(0.05, 0.95, 0.01, 100),
#regularization_rate=0.1,
#regularization_norm=l1,
#d_regularization_norm=d_l1,
)
y = y.astype(float)
# standartizing
x = (x - np.repeat(x.mean(axis=0), x.shape[0]).reshape((x.shape[0], x.mean(axis=0).shape[0]), order='F')) / \
np.sqrt(np.repeat(x.var(axis=0), x.shape[0]).reshape((x.shape[0], x.mean(axis=0).shape[0]), order='F'))
idx_train = list(np.random.choice(range(x.shape[0]), size=int(round(0.8 * x.shape[0]))))
idx_cv = list(set(range(x.shape[0])) - set(idx_train))
rbm = RBM(x.shape[1], 100,
rng=(lambda n: np.random.normal(0, 0.001, n)),
mode='gaus-bin')
print(rbm)
rbm.train(x[idx_train, :],
cd_k=1,
learning_rate=0.001,
momentum_rate=0.9,
max_iter=1000,
batch_size=20,
n_iter_stop_skip=10,
goal=euclidian,
#cv_input_data=cv_input,
stop_threshold=0.15,
#neural_local_gain=(0.05, 0.95, 0.01, 100),
regularization_rate=0.1,
#regularization_norm=l1,
d_regularization_norm=d_l1,
)
