def fit(self, train_x, train_y, val_x, val_y, bin_size, lr, batch_size,
with_gap, earlystop, verbose):
self.batch_size = batch_size
optimizer = torch.optim.Adam(self.model.parameters(),
lr=lr,
amsgrad=True)
loss_list = []
for i in range(self.maxepochs):
optimizer.zero_grad()
x, y = self.loader.get_batch()
x = x.to(self.device)
y = y.yo(self.device)
ypred = self.model(x)
if ypred.dim() == 2:
ypred = ypred.squeeze(1)
assert ypred.size() == y.size()
loss = MSELoss(reduction='mean')(ypred, y)
loss.backward()
optimizer.step()
if earlystop == True:
loss_list = loss_list.append(self.evaluate(val_x, val_y))
else:
if train_x.size()[0] == batch_size:
loss_list = loss_list.append(loss.cpu().data.numpy())
if len(loss_list)>5 \
and abs(loss_list[-2]/loss_list[-1]-1)<0.0001 :
break
if self.earlystop == True:
return None, loss_list[-1]
else:
return loss_list[-1], self.evaluate(val_x, val_y)
def compute_q_loss(global_state, state, reward, next_global_state,
next_state):
if args.cuda:
global_state = global_state.cuda()
state = state.cuda()
reward = reward.cuda()
next_global_state = next_global_state.cuda()
next_state = next_state.cuda()
global_state = Variable(global_state, requires_grad=True)
state = Variable(state, requires_grad=True)
next_global_state = Variable(next_global_state, volatile=True)
next_state = Variable(next_state, volatile=True)
current_q_values, _ = training_agents[0].act(global_state, state)
max_next_q_values, _ = training_agents[0].target_act(
next_global_state, next_state)
max_next_q_values = max_next_q_values.max(1)[0]
# sum the rewards for individual agents
expected_q_values = Variable(
reward.mean(dim=1)) + args.gamma * max_next_q_values
loss = MSELoss()(current_q_values, expected_q_values)
loss.backward()
return loss.cpu().data[0]
def inference_demo(screen):
forces = [hor_impulse]
ground_truth_mass = Variable(torch.DoubleTensor([7]))
world, c = make_world(forces, ground_truth_mass, num_links=NUM_LINKS)
rec = None
# rec = Recorder(DT, screen)
ground_truth_pos = positions_run_world(world, run_time=10, screen=screen, recorder=rec)
ground_truth_pos = [p.data for p in ground_truth_pos]
ground_truth_pos = Variable(torch.cat(ground_truth_pos))
learning_rate = 0.01
max_iter = 100
next_mass = Variable(torch.DoubleTensor([1.3]), requires_grad=True)
loss_hist = []
mass_hist = [next_mass]
last_dist = 1e10
for i in range(max_iter):
world, c = make_world(forces, next_mass, num_links=NUM_LINKS)
# world.load_state(initial_state)
# world.reset_engine()
positions = positions_run_world(world, run_time=10, screen=None)
positions = torch.cat(positions)
positions = positions[:len(ground_truth_pos)]
# temp_ground_truth_pos = ground_truth_pos[:len(positions)]
loss = MSELoss()(positions, ground_truth_pos)
loss.backward()
grad = c.mass.grad.data
# clip gradient
grad = torch.max(torch.min(grad, torch.DoubleTensor([100])), torch.DoubleTensor([-100]))
temp = c.mass.data - learning_rate * grad
temp = max(MASS_EPS, temp[0])
next_mass = Variable(torch.DoubleTensor([temp]), requires_grad=True)
# learning_rate /= 1.1
print(i, '/', max_iter, loss.data[0])
print(grad)
print(next_mass)
# print(learned_force(0.05))
if abs((last_dist - loss).data[0]) < 1e-3:
break
last_dist = loss
loss_hist.append(loss)
mass_hist.append(next_mass)
world = make_world(forces, next_mass, num_links=NUM_LINKS)[0]
# world.load_state(initial_state)
# world.reset_engine()
rec = None
# rec = Recorder(DT, screen)
positions_run_world(world, run_time=10, screen=screen, recorder=rec)
loss = MSELoss()(positions, ground_truth_pos)
print(loss.data[0])
print(next_mass)
plot(loss_hist)
plot(mass_hist)
def one_batch_train(self, Xs, Y):
Ws = Xs['word_vectors']
self.optimizer.zero_grad()
outputs = self.model(Ws)
mse_loss = MSELoss(outputs, Y)
mse_loss.backward()
self.optimizer.step()
self.save_checkpoint('../../../data/nn/lstm.pth.tar')
return self.model.data.cpu().numpy(), mse_loss.item()
def update(self, curr_states, next_states, actions, rewards, terminals,
discount, curr_model, eval_model, optimizer):
non_final_mask = 1 - torch.ByteTensor(terminals)
curr_states = Variable(curr_states, requires_grad=True)
next_states = Variable(next_states, requires_grad=False)
reward_batch = torch.FloatTensor(rewards)
actions = torch.LongTensor(actions).unsqueeze(1)
repeated_actions = actions.repeat(1,
len(self.reward_types)).unsqueeze(2)
# make prediction for decomposed q-values for current_state
decomp_q_curr_state, q_curr_state = curr_model(curr_states)
decomp_q_curr_state = decomp_q_curr_state.gather(2, repeated_actions)
q_curr_state = q_curr_state.gather(1, actions)
# make prediction for decomposed q-values for next_state
decomp_q_next_state, q_next_state = curr_model(
next_states[non_final_mask])
q_next_state, q_next_state_actions = q_next_state.max(1)
q_next_state_actions = q_next_state_actions.unsqueeze(1).repeat(
1, len(self.reward_types)).unsqueeze(2)
# Calculate Targets
target_q = Variable(torch.zeros(q_curr_state.shape),
requires_grad=False)
target_q[non_final_mask] = q_next_state.detach().unsqueeze(1)
target_q = discount * (reward_batch.sum(1, keepdim=True) + target_q)
target_decomp_q = Variable(torch.zeros(decomp_q_curr_state.shape),
requires_grad=False)
target_decomp_q[non_final_mask] = decomp_q_next_state.gather(
2, q_next_state_actions)
target_decomp_q = discount * (reward_batch.unsqueeze(2) +
target_decomp_q)
# calculate loss
loss = MSELoss()(q_curr_state, target_q)
loss += MSELoss()(decomp_q_curr_state, target_decomp_q)
optimizer.zero_grad()
loss.backward()
torch.nn.utils.clip_grad_norm_(curr_model.parameters(), 100)
optimizer.step()
def train_model():
model.train()
total_loss = 0
pbar = tqdm(train_loader)
for data in pbar:
if not args.multi_gpu: data = data.to(device)
optimizer.zero_grad()
#print(data)
#print(args)
if args.use_orig_graph:
x = data.x if args.use_feature else None
edge_weight = data.edge_weight if args.use_edge_weight else None
node_id = data.node_id if emb else None
out = model(data.z, data.edge_index, data.batch, x, edge_weight,
node_id).view(-1)
else:
out = model(data).view(-1)
if args.multi_gpu:
y = torch.cat([d.y.to(torch.float) for d in data]).to(out.device)
else:
y = data.y.to(torch.float)
if args.neg_edge_percent != 100:
y_neg = y[y == 0]
out_neg = out[y == 0]
y_pos = y[y != 0]
out_pos = out[y != 0]
num_neg = int(args.neg_edge_percent / 100 * len(out_neg))
out_neg, y_neg = out_neg[:num_neg], y_neg[:num_neg]
y = torch.cat([y_pos, y_neg])
out = torch.cat([out_pos, out_neg])
loss = MSELoss()(out, y)
loss = torch.sqrt(loss)
loss.backward()
optimizer.step()
total_loss += loss.item() * len(data)
return total_loss / len(train_dataset)
def main(screen):
forces = [hor_impulse]
ground_truth_mass = torch.tensor([TOTAL_MASS], dtype=DTYPE)
world, chain = make_world(forces, ground_truth_mass, num_links=NUM_LINKS)
rec = None
# rec = Recorder(DT, screen)
ground_truth_pos = positions_run_world(world, run_time=10, screen=screen, recorder=rec)
ground_truth_pos = [p.data for p in ground_truth_pos]
ground_truth_pos = torch.cat(ground_truth_pos)
learning_rate = 0.5
max_iter = 100
next_mass = torch.rand_like(ground_truth_mass, requires_grad=True)
print('\rInitial mass:', next_mass.item())
print('-----')
optim = torch.optim.RMSprop([next_mass], lr=learning_rate)
loss_hist = []
mass_hist = [next_mass.item()]
last_loss = 1e10
for i in range(max_iter):
if i % 1 == 0:
world, chain = make_world(forces, next_mass.clone().detach(), num_links=NUM_LINKS)
run_world(world, run_time=10, print_time=False, screen=None, recorder=None)
world, chain = make_world(forces, next_mass, num_links=NUM_LINKS)
positions = positions_run_world(world, run_time=10, screen=None)
positions = torch.cat(positions)
positions = positions[:len(ground_truth_pos)]
clipped_ground_truth_pos = ground_truth_pos[:len(positions)]
optim.zero_grad()
loss = MSELoss()(positions, clipped_ground_truth_pos)
loss.backward()
optim.step()
print('Iteration: {} / {}'.format(i+1, max_iter))
print('Loss:', loss.item())
print('Gradient:', next_mass.grad.item())
print('Next mass:', next_mass.item())
print('-----')
if abs((last_loss - loss).item()) < STOP_DIFF:
print('Loss changed by less than {} between iterations, stopping training.'
.format(STOP_DIFF))
break
last_loss = loss
loss_hist.append(loss.item())
mass_hist.append(next_mass.item())
world = make_world(forces, next_mass, num_links=NUM_LINKS)[0]
rec = None
positions = positions_run_world(world, run_time=10, screen=screen, recorder=rec)
positions = torch.cat(positions)
positions = positions[:len(ground_truth_pos)]
clipped_ground_truth_pos = ground_truth_pos[:len(positions)]
loss = MSELoss()(positions, clipped_ground_truth_pos)
print('Final loss:', loss.item())
print('Final mass:', next_mass.item())
plot(loss_hist)
plot(mass_hist)
def main(screen):
dateTimeObj = datetime.now()
timestampStr = dateTimeObj.strftime("%d-%b-%Y(%H:%M)")
print('Current Timestamp : ', timestampStr)
if not os.path.isdir(timestampStr):
os.mkdir(timestampStr)
#if torch.cuda.is_available():
# dev = "cuda:0"
#else:
# dev = "cpu"
forces = []
rec = None
#rec = Recorder(DT, screen)
#plot(zhist)
learning_rate = 0.5
max_iter = 100
plt.subplots(21)
plt.subplot(212)
plt.gca().invert_yaxis()
optim = torch.optim.RMSprop([utT], lr=learning_rate)
last_loss = 1e10
lossHist = []
for i in range(1, 20000):
world, chain = make_world(forces, ctT, utT)
hist = positions_run_world(world,
run_time=runtime,
screen=screen,
recorder=rec)
xy = hist[0]
vel = hist[1]
control = hist[2]
xy = torch.cat(xy).to(device=Defaults.DEVICE)
x1 = xy[0::2]
y1 = xy[1::2]
control = torch.cat(control).to(device=Defaults.DEVICE)
optim.zero_grad()
targetxy = []
targetControl = []
j = 0
while (j < xy.size()[0]):
targetxy.append(500)
targetxy.append(240)
targetControl.append(0)
targetControl.append(0)
j = j + 2
tt = torch.tensor(targetxy,
requires_grad=True,
dtype=DTYPE,
device=Defaults.DEVICE).t()
tc = torch.tensor(targetControl,
requires_grad=True,
dtype=DTYPE,
device=Defaults.DEVICE).t()
#loss = MSELoss()(zhist, 150*torch.tensor(np.ones(zhist.size()),requires_grad=True, dtype=DTYPE,device=Defaults.DEVICE))/100
#loss = MSELoss()(xy, tt) / 10 + 0*MSELoss()(control, tc) / 1000 + abs(vel[-1][0]) + abs(vel[-1][1]) + abs(vel[-1][2])
loss = MSELoss()(xy, tt) / 10 + 0 * MSELoss()(
control, tc) / 1000 + abs(vel[-1][0]) + abs(vel[-1][1]) + abs(
vel[-1][2])
#loss = zhist[-1]
loss.backward()
lossHist.append(loss.item())
optim.step()
print('Loss:', loss.item())
print('Gradient:', utT.grad)
print('Next u:', utT)
#plt.axis([-1, 1, -1, 1])
plt.ion()
plt.show()
plt.subplot(211)
plt.plot(lossHist)
plt.draw()
pl1 = xy.cpu()
plt.subplot(212)
plt.plot(pl1.clone().detach().numpy()[::4],
pl1.clone().detach().numpy()[1::4])
#plt.plot(zhist)
plt.draw()
plt.pause(0.001)
plt.savefig('2step.png')
if (i % 20) == 0:
plt.subplot(211)
plt.cla()
plt.subplot(212)
plt.cla()
plt.gca().invert_yaxis()
lossHist.clear()
torch.save(utT, timestampStr + '/file' + str(i) + '.pt')
utTCurrent = utT.clone()
world, chain = make_world(forces, ctT, utTCurrent)
positions_run_world(world, run_time=runtime, screen=None, recorder=rec)
def train(self, epochs, load=False):
print('TRAINING MODEL:'
' BATCH_SIZE = ' + str(Trainer.BATCH_SIZE) + ', PARTICLE_DIM: ' +
str(PARTICLE_DIM) + ', EPOCHS: ' + str(epochs) +
', PRTCL_LATENT_SPACE_SIZE: ' +
str(self.PRTCL_LATENT_SPACE_SIZE))
encoder, decoder, discriminator = self.create_model()
enc_optim = torch.optim.Adam(encoder.parameters(), lr=self.LR)
dec_optim = torch.optim.Adam(decoder.parameters(), lr=self.LR)
dis_optim = torch.optim.Adam(discriminator.parameters(), lr=self.LR)
embedder = self.create_embedder()
deembedder = self.create_deembedder()
if load:
print('LOADING MODEL STATES...')
encoder.load_state_dict(torch.load(Trainer.ENCODER_SAVE_PATH))
decoder.load_state_dict(torch.load(Trainer.DECODER_SAVE_PATH))
discriminator.load_state_dict(
torch.load(Trainer.DISCRIMINATOR_SAVE_PATH))
embedder.load_state_dict(torch.load(Trainer.PDG_EMBED_SAVE_PATH))
deembedder.load_state_dict(
torch.load(Trainer.PDG_DEEMBED_SAVE_PATH))
print('AUTOENCODER')
print(encoder)
print(decoder)
print(discriminator)
print('EMBEDDER')
print(embedder)
print('DEEMBEDDER')
print(deembedder)
_data = load_data()
data_train, data_valid = self.prep_data(_data,
batch_size=Trainer.BATCH_SIZE,
valid=0.1)
particles = torch.tensor(particle_idxs(), device=self.device)
particles.requires_grad = False
for epoch in range(epochs):
for n_batch, batch in enumerate(data_train):
encoder.zero_grad()
decoder.zero_grad()
discriminator.zero_grad()
real_data: torch.Tensor = batch.to(self.device)
emb_data = self.embed_data(real_data, [embedder]).detach()
batch_size = len(batch)
zeros = torch.zeros(batch_size,
device=self.device,
requires_grad=False)
ones = torch.ones(batch_size,
device=self.device,
requires_grad=False)
# ======== Train Discriminator ======== #
decoder.freeze(True)
encoder.freeze(True)
discriminator.freeze(False)
lat_fake = torch.randn(batch_size,
self.PRTCL_LATENT_SPACE_SIZE,
device=self.device)
disc_fake = discriminator(lat_fake)
lat_real = encoder(emb_data)
disc_real = discriminator(lat_real)
loss_fake = MSELoss()(disc_fake, zeros)
loss_real = MSELoss()(disc_real, ones)
loss_fake.backward()
loss_real.backward()
dis_optim.step()
# ======== Train Generator ======== #
decoder.freeze(False)
encoder.freeze(False)
discriminator.freeze(True)
lat_real = encoder(emb_data)
recon_data = decoder(lat_real)
d_real = discriminator(encoder(emb_data))
recon_loss = MSELoss()(emb_data, recon_data)
d_loss = MSELoss()(d_real, zeros)
recon_loss.backward()
d_loss.backward()
enc_optim.step()
dec_optim.step()
self.train_deembeders([
(particles, embedder, deembedder),
],
epochs=2)
if n_batch % 100 == 0:
self.print_deemb_quality(particles, embedder, deembedder)
self.show_heatmaps(emb_data[:30, :],
recon_data[:30, :],
reprod=False,
save=True,
epoch=epoch,
batch=n_batch)
err_kld, err_wass = self.gen_show_comp_hists(
decoder,
_data,
attr_idxs=[
FEATURES - 8, FEATURES - 7, FEATURES - 6,
FEATURES - 5
],
embedders=[embedder],
emb=False,
deembedder=deembedder,
save=True,
epoch=epoch,
batch=n_batch)
self.errs_kld.append(err_kld)
self.errs_wass.append(err_wass)
valid_loss = self._valid_loss(encoder, decoder, embedder,
data_valid)
print(
f'Epoch: {str(epoch)}/{epochs} :: '
f'Batch: {str(n_batch)}/{str(len(data_train))} :: '
f'train loss: {"{:.6f}".format(round(recon_loss.item(), 6))} :: '
f'valid loss: {"{:.6f}".format(round(valid_loss, 6))} :: '
f'err kld: {"{:.6f}".format(round(err_kld, 6))} :: '
f'err wass: {"{:.6f}".format(round(err_wass, 6))}')
self._save_models(encoder, decoder, discriminator, embedder,
deembedder)
with open(self.ERRS_SAVE_PATH, 'wb') as handle:
pickle.dump((self.errs_kld, self.errs_wass),
handle,
protocol=pickle.HIGHEST_PROTOCOL)
return encoder, decoder, discriminator, embedder, deembedder
def train(*args, **kwargs):
print(kwargs)
device = kwargs['train']['device']
# 1. Init audio preprocessing
preproc = AudioPreprocessor(**kwargs['preprocessing_params'])
sr = kwargs['preprocessing_params']['sample_rate']
# 2. Load preprocessing net
preproc_net = torch.load(kwargs['preproc_net_fname']).to(device)
# 3. Init model dynamics net
md_net = LstmModelDynamics(**kwargs['model_dynamics_params']).to(device)
optim = torch.optim.Adam(md_net.parameters(), lr=kwargs['train']['learning_rate'], eps=kwargs['train']['learning_rate_eps'])
# 4. Load training set
data_fname = kwargs['data_fname']
df = pd.read_pickle(data_fname)
# 5. Train loop
params = kwargs['train']
md_net.train()
for i in range(params['num_steps']):
sample = df.sample(n=kwargs['train']['minibatch_size'])
states = np.stack(sample.loc[:, 'states'].values)
actions = np.stack(sample.loc[:, 'actions'].values)
audio = np.stack(sample.loc[:, 'audio'].values)
preproc_audio = np.array([preproc(audio[j], sr) for j in range(audio.shape[0])])
acoustic_states = torch.from_numpy(preproc_audio).float().to(device)
# acoustic_states = acoustic_states.view(-1, kwargs['model_dynamics_params']["acoustic_state_dim"])
# mean_norm = acoustic_states.mean(dim=0)
# mean_std = acoustic_states.std(dim=0)
# acoustic_states = (acoustic_states - mean_norm.view(1, -1)) / mean_std.view(1, -1)
# acoustic_states = acoustic_states.view(kwargs['train']['minibatch_size'], -1, kwargs['model_dynamics_params']["acoustic_state_dim"])
_, _, acoustic_states = preproc_net(torch.from_numpy(preproc_audio).float().to(device),
seq_lens=np.array([preproc_audio.shape[-2]]))
seq_len = actions.shape[1]
acoustic_state_dim = kwargs['model_dynamics_params']["acoustic_state_dim"]
# forward prop
lstm_outs, predicted_acoustic_states = md_net(acoustic_states,
torch.from_numpy(states[:, :seq_len, :]).float().to(device),
torch.from_numpy(actions).float().to(device))
# compute error
loss = MSELoss(reduction='sum')(predicted_acoustic_states[:, :-1, :].contiguous().view(-1, acoustic_state_dim),
acoustic_states[:, 1:, :].contiguous().view(-1, acoustic_state_dim)) / (seq_len * kwargs['train']['minibatch_size'])
# backprop
optim.zero_grad()
loss.backward()
optim.step()
dynamics = MSELoss(reduction='sum')(acoustic_states[:, :-1, :].contiguous().view(-1, acoustic_state_dim),
acoustic_states[:, 1:, :].contiguous().view(-1, acoustic_state_dim)) / (seq_len * kwargs['train']['minibatch_size'])
print("\rstep: {} | loss: {:.4f}| actual_dynamics: {:.4f}".format(i, loss.detach().cpu().item(), dynamics.detach().cpu().item()), end="")
def train(self, epochs, load=False):
print('TRAINING MODEL:'
' BATCH_SIZE = ' + str(self.BATCH_SIZE) + ', PARTICLE_DIM: ' +
str(PARTICLE_DIM) + ', EPOCHS: ' + str(epochs) +
', PRTCL_LATENT_SPACE_SIZE: ' +
str(self.PRTCL_LATENT_SPACE_SIZE))
generator, discriminator = self.create_model()
gen_optim = torch.optim.Adam(generator.parameters(),
lr=self.LR,
betas=(0, .9))
dis_optim = torch.optim.Adam(discriminator.parameters(),
lr=self.LR,
betas=(0, .9))
embedder = self.create_embedder()
deembedder = self.create_deembedder()
particles = torch.tensor(particle_idxs(), device=self.device)
if load:
print('LOADING MODEL STATES...')
generator.load_state_dict(torch.load(Trainer.GENERATOR_SAVE_PATH))
discriminator.load_state_dict(
torch.load(Trainer.DISCRIMINATOR_SAVE_PATH))
embedder.load_state_dict(torch.load(Trainer.PDG_EMBED_SAVE_PATH))
deembedder.load_state_dict(
torch.load(Trainer.PDG_DEEMBED_SAVE_PATH))
print('GENERATOR')
print(generator)
print('DISCRIMINATOR')
print(discriminator)
print('EMBEDDER')
print(embedder)
print('DEEMBEDDER')
print(deembedder)
_data = load_data()
data_train, data_valid = self.prep_data(_data,
batch_size=self.BATCH_SIZE,
valid=0.1)
for epoch in range(epochs):
for n_batch, batch in enumerate(data_train):
real_data: torch.Tensor = batch.to(self.device)
emb_data = self.embed_data(real_data, [embedder]).detach()
batch_size = len(batch)
valid = torch.ones(batch_size,
device=self.device,
requires_grad=False)
fake = torch.zeros(batch_size,
device=self.device,
requires_grad=False)
# ======== Train Generator ======== #
gen_optim.zero_grad()
# Sample noise as generator input
lat_fake = torch.randn(batch_size,
self.PRTCL_LATENT_SPACE_SIZE,
device=self.device)
lat_fake = Variable(
torch.tensor(np.random.normal(
0, 1, (batch_size, self.PRTCL_LATENT_SPACE_SIZE)),
device=self.device).float())
# Generate a batch of images
gen_data = generator(lat_fake)
# Loss measures generator's ability to fool the discriminator
g_loss = MSELoss()(discriminator(gen_data), valid)
g_loss.backward()
gen_optim.step()
# ======== Train Discriminator ======== #
dis_optim.zero_grad()
real_loss = MSELoss()(discriminator(emb_data), valid)
fake_loss = MSELoss()(discriminator(gen_data.detach()), fake)
d_loss = (real_loss + fake_loss) / 2
d_loss.backward()
dis_optim.step()
self.train_deembeders([
(particles, embedder, deembedder),
],
epochs=2)
if n_batch % 100 == 0:
self.print_deemb_quality(
torch.tensor(particle_idxs(), device=self.device),
embedder, deembedder)
self.show_heatmaps(emb_data[:30, :],
gen_data[:30, :],
reprod=False,
save=True,
epoch=epoch,
batch=n_batch)
err_kld, err_wass = self.gen_show_comp_hists(
generator,
_data,
attr_idxs=[
FEATURES - 8, FEATURES - 7, FEATURES - 6,
FEATURES - 5
],
embedders=[embedder],
emb=False,
deembedder=deembedder,
save=True,
epoch=epoch,
batch=n_batch)
self.errs_kld.append(err_kld)
self.errs_wass.append(err_wass)
print(
f'Epoch: {str(epoch)}/{epochs} :: '
f'Batch: {str(n_batch)}/{str(len(data_train))} :: '
f'generator loss: {"{:.6f}".format(round(g_loss.item(), 6))} :: '
f'discriminator loss: {"{:.6f}".format(round(d_loss.item(), 6))} :: '
f'err kld: {"{:.6f}".format(round(err_kld, 6))} :: '
f'err wass: {"{:.6f}".format(round(err_wass, 6))}')
self._save_models(generator, discriminator, embedder, deembedder)
with open(self.ERRS_SAVE_PATH, 'wb') as handle:
pickle.dump((self.errs_kld, self.errs_wass),
handle,
protocol=pickle.HIGHEST_PROTOCOL)
return generator, discriminator, embedder, deembedder