G_X,
data.dim_X,
Z,
)
view_dist = view_dist_setup(X, Z)
view_disentangle = view_disentangle_setup(
X,
G_X,
data.dim_X,
Z,
)
sess = tf.Session()
sess.run(tf.global_variables_initializer())
data_feed = default_feeder(data, X, mb_size)
print(sess.run([recon_loss, tf.reduce_mean(kl_losses)], feed_dict=data_feed))
run_ae(data=data,
mb_size=mb_size,
interpolation=interpolation,
feature_eval=None,
view_dist=view_dist,
view_disentangle=view_disentangle,
train=train,
loss=loss,
X=X,
G_X=G_X,
sess=sess,
experiment_id=name,
max_iter=10000000)
train_G = tf.train.AdamOptimizer(lr).minimize(recon_loss,
var_list=vars_G + vars_E)
train_E = tf.train.AdamOptimizer(lr).minimize(G_loss, var_list=vars_E)
print([_.name for _ in tf.get_collection(tf.GraphKeys.GLOBAL_VARIABLES)])
train = tf.group(train_D, train_E, train_G)
sess = tf.Session()
view_dist = view_dist_setup(X, Z)
view_disentangle = view_disentangle_setup(
X,
G_X,
data.dim_X,
Z,
)
sess.run(tf.global_variables_initializer())
data_feed = default_feeder(data, X, mb_size)
run_ae(data=data,
mb_size=mb_size,
interpolation=None,
view_disentangle=view_disentangle,
view_dist=view_dist,
feature_eval=None,
train=train,
loss=recon_loss,
X=X,
G_X=G_X,
sess=sess,
experiment_id=name)
tf.reduce_sum((Xt_mean - X__)**2, axis=(1, 2, 3)))
recon_loss += recon_loss_t
G_X = X_means[0]
loss = recon_loss + kl_loss / T
train = tf.train.AdamOptimizer(learning_rate=1e-4).minimize(loss)
print([_.name for _ in tf.get_collection(tf.GraphKeys.GLOBAL_VARIABLES)])
sess = tf.Session()
interpolation = interpolation_setup(
X,
G_X,
data.dim_X,
Zs[0],
)
sess.run(tf.global_variables_initializer())
data_feed = default_feeder(data, X, mb_size)
print(sess.run([recon_loss, kl_loss], feed_dict=data_feed))
run_ae(data=data,
mb_size=mb_size,
interpolation=interpolation,
feature_eval=None,
train=train,
loss=loss,
X=X,
G_X=G_X,
sess=sess,
experiment_id=name)
with tf.variable_scope('G'):
# G_logits = nets.dense_net(Z, [256, data.dim_x], batch_norm=True)
G_logits = nets.deconv64(Z)
G_X = tf.nn.sigmoid(G_logits)
loss = tf.reduce_mean(
tf.nn.sigmoid_cross_entropy_with_logits(logits=G_logits, labels=X__))
train = tf.train.RMSPropOptimizer(learning_rate=1e-4).minimize(loss)
print([_.name for _ in tf.get_collection(tf.GraphKeys.GLOBAL_VARIABLES)])
sess = tf.Session()
feature_eval = feature_eval_setup(sess,
X,
Z,
data.sample(1000),
data_test.sample(1000),
accuracy,
nets.smcewl,
max_iter=1000)
sess.run(tf.global_variables_initializer())
run_ae(data=data,
mb_size=mb_size,
feature_eval=feature_eval,
train=train,
loss=loss,
X=X,
G_X=G_X,
sess=sess,
experiment_id=name)