def _objective(self):
n_train_examples = 50000
train_x_l, train_l_y, train_u_x, train_u_y, self.valid_x, self.valid_y, self.test_x, self.test_y = extract_data(
n_train_examples)
self.train_x = np.concatenate((train_x_l, train_u_x), axis=0)
self.train_y = np.concatenate((train_l_y, train_u_y), axis=0)
num_batches = int(n_train_examples / self.batch_size)
logging.debug("num batches:{}, batch_size:{}, epochs:{}".format(num_batches, self.batch_size, int(
self.num_iterations / num_batches)))
self.y_logits, self.y_pred_cls, self.cost = self.build_model()
tf.summary.scalar('cost', self.cost)
self.optimizer = tf.train.AdamOptimizer(learning_rate=self.learning_rate, beta1=self.beta1,
beta2=self.beta2).minimize(self.cost)
def _objective(self):
n_train_examples = 50000
train_x_l, train_l_y, train_u_x, train_u_y, self.valid_x, self.valid_y, self.test_x, self.test_y = extract_data(
n_train_examples)
num_batches = int(n_train_examples / self.batch_size)
logging.debug("num batches:{}, batch_size:{}, epochs:{}".format(
num_batches, self.batch_size,
int(self.num_iterations / num_batches)))
self.train_x = np.concatenate((train_x_l, train_u_x), axis=0)
self.train_y = np.concatenate((train_l_y, train_u_y), axis=0)
elbo, self.x_recon_mu, self.z_sample, self.z_mu, self.z_logvar, self.loglik = self.build_model(
)
self.cost = (elbo * num_batches +
prior_weights()) / (-self.batch_size * num_batches)
self.optimizer = tf.train.AdamOptimizer(
learning_rate=self.learning_rate,
beta1=self.beta1,
beta2=self.beta2).minimize(self.cost)
'n_labeled': 50000,
'alpha': 1, # 0.1 - 2 TODO change alpha back to 0.1
'latent_dim': 22, # should be 50 TODO change back to 50
'require_improvement': 5000,
'n_train': 50000,
'learning_rate': 3e-4,
'beta1': 0.9,
'beta2': 0.999,
'input_dim': 28 * 28,
'num_classes': 10,
'min_std':
0.1, # Dimensions with std < min_std are removed before training with GC
'l2_weight': 1e-6
}
train_x_lab, train_l_y, train_x_unlab, train_u_y, valid_x, valid_y, test_x, test_y = extract_data(
FLAGS['n_labeled'])
train_x_l_mu, train_x_l_logvar, train_x_u_mu, train_x_u_logvar, valid_x_mu, \
valid_x_logvar, test_x_mu, test_x_logvar = encode_dataset(FLAGS=FLAGS, train_lab=train_x_lab,
train_unlab=train_x_unlab, valid=valid_x,
test=test_x, min_std=FLAGS['min_std'])
train_lab = [train_x_l_mu, train_x_l_logvar, train_l_y]
train_unlab = [train_x_u_mu, train_x_u_logvar, train_u_y]
valid = [valid_x_mu, valid_x_logvar, valid_y]
test = [test_x_mu, test_x_logvar, test_y]
print("train lab: mu {}, var:{}, y:{}".format(train_x_l_mu.shape,
train_x_l_logvar.shape,
train_l_y.shape))
print("train unlab: mu {}, var:{}, y:{}".format(train_x_u_mu.shape,
train_x_u_logvar.shape,
train_u_y.shape))
if __name__ == '__main__':
FLAGS = {
'num_iterations': 40000, # should 3000 epochs
'batch_size': 200,
'seed': 31415,
'alpha': 0.1,
'require_improvement': 5000,
'n_train': 50000,
'learning_rate': 3e-4,
'beta1': 0.9,
'beta2': 0.999,
'num_classes': 10,
'n_components': 22
}
train_x_l, train_l_y, train_u_x, train_u_y, valid_x, valid_y, test_x, test_y = extract_data(
FLAGS['n_train'])
train_x = np.concatenate((train_x_l, train_u_x), axis=0)
train_y = np.concatenate((train_l_y, train_u_y), axis=0)
pca = pca_components(train_x, FLAGS['n_components'])
train_x_tran = pca.transform(train_x)
train = [train_x_tran, train_y]
valid_x_tran = pca.transform(valid_x)
valid = [valid_x_tran, valid_y]
test_x_tran = pca.transform(test_x)
test = [test_x_tran, test_y]
print("train_x_tran:{}, valid_x_tran:{}, test_x_tran:{} ".format(