def fit(self, sess, local_):
for _ in range(local_):
x, _ = next_batch_(FLAGS.bz)
sess.run(self.optim, {self.x: x})
x, _ = next_batch_(FLAGS.bz * 5)
return sess.run(
[self.loss, self.rec_loss, self.kld_loss, self.fit_summary],
{self.x: x})
def fit(self, sess, local_):
for _ in range(local_):
x, _ = next_batch_(FLAGS.bz)
sess.run(self.optim, {
self.x: x,
self.z: gaussian(FLAGS.bz, FLAGS.z_dim)
})
x, _ = next_batch_(FLAGS.bz * 5)
return sess.run(
[self.loss, self.loss_nll, self.loss_mmd, self.fit_summary], {
self.x: x,
self.z: gaussian(FLAGS.bz, FLAGS.z_dim)
})
def fit(self, sess, local_):
for _ in range(local_):
x_real, y = next_batch_(FLAGS.bz)
one_hot_y = one_hot_(y, FLAGS.y_dim)
sess.run(self.a_optim, {self.x: x_real})
for _ in range(3):
sess.run(self.d_optim, {self.x: x_real, self.y: one_hot_y, self.real_z: z_real_(y)})
sess.run(self.g_optim, {self.x: x_real, self.y: one_hot_y})
x_real, y = next_batch_(FLAGS.bz * 5)
one_hot_y = one_hot_(y, FLAGS.y_dim)
return sess.run([self.a_loss, self.g_loss, self.d_loss, self.fit_summary], {
self.x: x_real, self.real_z: z_real_(y), self.y: one_hot_y})
def fit(self, sess, local_):
for _ in range(local_):
x_real, _ = next_batch_(FLAGS.bz)
sess.run(self.a_optim, {self.x: x_real})
for _ in range(3):
sess.run(self.d_optim, {
self.x: x_real,
self.real_z: z_real_(FLAGS.bz)
})
sess.run(self.g_optim, {self.x: x_real})
x_real, _ = next_batch_(FLAGS.bz * 5)
return sess.run(
[self.a_loss, self.g_loss, self.d_loss, self.fit_summary], {
self.x: x_real,
self.real_z: z_real_(FLAGS.bz * 5)
})
def latent_z(self, sess, bz):
x, y = next_batch_(bz)
return sess.run(self.fake_z, {self.x: x}), y