var_list=para_G)
solver_KSD_gp_p = optimizer_KSD(learning_rate=lr_p).minimize(KSD_gp,
var_list=para_P)
#----------------------------- training ------------------------------------
sess = tf.Session()
if load_it == 0:
sess.run(tf.global_variables_initializer())
else:
saver = tf.train.Saver()
saver.restore(sess, save_path=workdir + '/model/{}'.format(load_it))
best_mmd, best_kl, best_rate = 10.0, 10.0, 0.
datainput = DataInput(data=true_sample, batch_size=mb_size)
start_time = datetime.now()
iterations = 0
for it in range(load_it, load_it + n_iter + 1):
i_critic = 0
while i_critic <= n_critic:
_, loss_d = sess.run(
[solver_WGAN_D, D_loss],
feed_dict={
z: sample_z(mb_size, z_dim),
X_true: datainput.get_batch(),
isTraining: True
})
fake_sample = sess.run(G_sample,
feed_dict={
z: sample_z(show_size, z_dim),
isTraining: False
})
fake_density = sess.run(Density,
feed_dict={
Mesh: fake_sample,
isTraining: False
})
X_range, Y_range = XY_range[0], XY_range[1]
mesh_input = np.stack([np.reshape(X_range, [-1]),
np.reshape(Y_range, [-1])], 1)
meshinput = DataInput(data=mesh_input, batch_size=mb_size)
mesh_f_density = np.zeros([mesh_input.shape[0], 1])
for i in range(mesh_input.shape[0] // mb_size):
mesh_f_density[i * mb_size:min((i + 1) *
mb_size, mesh_input.shape[0])] = sess.run(
Density,
feed_dict={
Mesh: meshinput.get_batch(),
isTraining: False
})
mesh_f_density = np.reshape(mesh_f_density,
[X_range.shape[0], X_range.shape[1]])
norm = np.sum(mesh_t_density)
mesh_f_density_n = mesh_f_density / np.sum(mesh_f_density) * norm
solver_KSD_gp_g = optimizer_KSD(learning_rate=lr_g).minimize(KSD_gp, var_list=para_G)
solver_KSD_gp_p = optimizer_KSD(learning_rate=lr_p).minimize(KSD_gp, var_list=para_P)
#----------------------------- training ------------------------------------
sess = tf.Session()
if load_it==0:
sess.run(tf.global_variables_initializer())
else:
saver = tf.train.Saver()
saver.restore(sess, save_path=workdir+'/model/{}'.format(load_it))
best_mmd, best_kl, best_rate = 10.0, 10.0, 0.
datainput = DataInput(data=true_sample, batch_size=mb_size)
start_time = datetime.now()
iterations = 0
for it in range(load_it, load_it+n_iter+1):
i_critic = 0
while i_critic<=n_critic:
_, loss_d = sess.run([solver_GAN_D, D_loss],
feed_dict={z: sample_z(mb_size, z_dim),
X_true: datainput.get_batch(),
isTraining: True})
i_critic += 1
_, loss_g = sess.run([solver_GAN_G, G_loss],