def get_grads_correct(seed):
util.set_seed(seed)
theta_grads_correct = []
phi_grads_correct = []
log_weight, log_q = losses.get_log_weight_and_log_q(
generative_model, inference_network, obs, num_particles)
optimizer_phi.zero_grad()
optimizer_theta.zero_grad()
wake_theta_loss, elbo = losses.get_wake_theta_loss_from_log_weight(
log_weight)
wake_theta_loss.backward(retain_graph=True)
theta_grads_correct = [parameter.grad.clone() for parameter in
generative_model.parameters()]
# in rws, we step as we compute the grads
# optimizer_theta.step()
optimizer_phi.zero_grad()
optimizer_theta.zero_grad()
wake_phi_loss = losses.get_wake_phi_loss_from_log_weight_and_log_q(
log_weight, log_q)
wake_phi_loss.backward()
phi_grads_correct = [parameter.grad.clone() for parameter in
inference_network.parameters()]
# in rws, we step as we compute the grads
# optimizer_phi.step()
return theta_grads_correct, phi_grads_correct
def get_grads_in_one_no_zeroing(seed):
util.set_seed(seed)
theta_grads_in_one = []
phi_grads_in_one = []
log_weight, log_q = losses.get_log_weight_and_log_q(
generative_model, inference_network, obs, num_particles)
optimizer_phi.zero_grad()
optimizer_theta.zero_grad()
wake_theta_loss, elbo = losses.get_wake_theta_loss_from_log_weight(
log_weight)
wake_theta_loss.backward(retain_graph=True)
# optimizer_phi.zero_grad() -> don't zero phi grads
# optimizer_theta.zero_grad()
wake_phi_loss = losses.get_wake_phi_loss_from_log_weight_and_log_q(
log_weight, log_q)
wake_phi_loss.backward()
# only get the grads in the end!
theta_grads_in_one = [parameter.grad.clone() for parameter in
generative_model.parameters()]
phi_grads_in_one = [parameter.grad.clone() for parameter in
inference_network.parameters()]
# in pyro, we want step to be in a different stage
# optimizer_theta.step()
# optimizer_phi.step()
return theta_grads_in_one, phi_grads_in_one
def get_grads_correct_sleep(seed):
util.set_seed(seed)
theta_grads_correct = []
phi_grads_correct = []
log_weight, log_q = losses.get_log_weight_and_log_q(
generative_model, inference_network, obs, num_particles)
optimizer_phi.zero_grad()
optimizer_theta.zero_grad()
wake_theta_loss, elbo = losses.get_wake_theta_loss_from_log_weight(
log_weight)
wake_theta_loss.backward(retain_graph=True)
theta_grads_correct = [
parameter.grad.clone() for parameter in generative_model.parameters()
]
# in rws, we step as we compute the grads
# optimizer_theta.step()
optimizer_phi.zero_grad()
optimizer_theta.zero_grad()
wake_phi_loss = losses.get_wake_phi_loss_from_log_weight_and_log_q(
log_weight, log_q)
wake_phi_loss.backward()
wake_phi_grads_correct = [
parameter.grad.clone() for parameter in inference_network.parameters()
]
# in rws, we step as we compute the grads
# optimizer_phi.step()
optimizer_phi.zero_grad()
optimizer_theta.zero_grad()
sleep_phi_loss = losses.get_sleep_loss(generative_model,
inference_network,
num_samples=num_particles)
sleep_phi_loss.backward()
sleep_phi_grads_correct = [
parameter.grad.clone() for parameter in inference_network.parameters()
]
wake_factor = 0.755
phi_grads_correct = [
wake_factor * wake_phi_grad_correct +
(1 - wake_factor) * sleep_phi_grad_correct
for wake_phi_grad_correct, sleep_phi_grad_correct in zip(
wake_phi_grads_correct, sleep_phi_grads_correct)
]
return theta_grads_correct, phi_grads_correct
def train_wake_wake(generative_model,
inference_network,
data_loader,
num_iterations,
num_particles,
optim_kwargs,
callback=None):
optimizer_phi = torch.optim.Adam(inference_network.parameters(),
**optim_kwargs)
optimizer_theta = torch.optim.Adam(generative_model.parameters(),
**optim_kwargs)
iteration = 0
while iteration < num_iterations:
for obs in iter(data_loader):
log_weight, log_q = losses.get_log_weight_and_log_q(
generative_model, inference_network, obs, num_particles)
# wake theta
optimizer_phi.zero_grad()
optimizer_theta.zero_grad()
wake_theta_loss, elbo = losses.get_wake_theta_loss_from_log_weight(
log_weight)
wake_theta_loss.backward(retain_graph=True)
optimizer_theta.step()
# wake phi
optimizer_phi.zero_grad()
optimizer_theta.zero_grad()
wake_phi_loss = losses.get_wake_phi_loss_from_log_weight_and_log_q(
log_weight, log_q)
wake_phi_loss.backward()
optimizer_phi.step()
if callback is not None:
callback(iteration,
wake_theta_loss.item(), wake_phi_loss.item(),
elbo.item(), generative_model, inference_network,
optimizer_theta, optimizer_phi)
iteration += 1
# by this time, we have gone through `iteration` iterations
if iteration == num_iterations:
break
def train_defensive_wake_wake(delta,
generative_model,
inference_network,
obss_data_loader,
num_iterations,
num_particles,
callback=None):
optimizer_phi = torch.optim.Adam(inference_network.parameters())
optimizer_theta = torch.optim.Adam(generative_model.parameters())
obss_iter = iter(obss_data_loader)
for iteration in range(num_iterations):
# get obss
try:
obss = next(obss_iter)
except StopIteration:
obss_iter = iter(obss_data_loader)
obss = next(obss_iter)
log_weight, log_q = losses.get_log_weight_and_log_q(
generative_model, inference_network, obss, num_particles)
# wake theta
optimizer_phi.zero_grad()
optimizer_theta.zero_grad()
wake_theta_loss, elbo = losses.get_wake_theta_loss_from_log_weight(
log_weight)
wake_theta_loss.backward(retain_graph=True)
optimizer_theta.step()
# wake phi
optimizer_phi.zero_grad()
optimizer_theta.zero_grad()
wake_phi_loss = losses.get_defensive_wake_phi_loss(
generative_model, inference_network, obss, delta, num_particles)
wake_phi_loss.backward()
optimizer_phi.step()
if callback is not None:
callback(iteration, wake_theta_loss.item(), wake_phi_loss.item(),
elbo.item(), generative_model, inference_network,
optimizer_theta, optimizer_phi)
return optimizer_theta, optimizer_phi
def get_log_p_and_kl(generative_model, inference_network, obs, num_particles):
"""Compute log weight and log prob of inference network.
Args:
generative_model: models.GenerativeModel object
inference_network: models.InferenceNetwork object
obs: tensor of shape [batch_size, num_data * num_dim]
num_particles: int
Returns:
log_p: tensor [batch_size]
kl: tensor [batch_size]
"""
log_weight, _ = losses.get_log_weight_and_log_q(generative_model,
inference_network, obs,
num_particles)
log_p = torch.logsumexp(log_weight, dim=1) - math.log(num_particles)
elbo = torch.mean(log_weight, dim=1)
kl = log_p - elbo
return log_p, kl
def train_wake_wake(generative_model,
inference_network,
true_generative_model,
batch_size,
num_iterations,
num_particles,
callback=None):
optimizer_phi = torch.optim.Adam(inference_network.parameters())
optimizer_theta = torch.optim.Adam(generative_model.parameters())
for iteration in range(num_iterations):
# generate synthetic data
obss = [true_generative_model.sample_obs() for _ in range(batch_size)]
log_weight, log_q = losses.get_log_weight_and_log_q(
generative_model, inference_network, obss, num_particles)
# wake theta
optimizer_phi.zero_grad()
optimizer_theta.zero_grad()
wake_theta_loss, elbo = losses.get_wake_theta_loss_from_log_weight(
log_weight)
wake_theta_loss.backward(retain_graph=True)
optimizer_theta.step()
# wake phi
optimizer_phi.zero_grad()
optimizer_theta.zero_grad()
wake_phi_loss = losses.get_wake_phi_loss_from_log_weight_and_log_q(
log_weight, log_q)
wake_phi_loss.backward()
optimizer_phi.step()
if callback is not None:
callback(iteration, wake_theta_loss.item(), wake_phi_loss.item(),
elbo.item(), generative_model, inference_network,
optimizer_theta, optimizer_phi)
return optimizer_theta, optimizer_phi
def get_mean_stds(generative_model, inference_network, num_mc_samples, obss,
num_particles):
vimco_grad = util.OnlineMeanStd()
vimco_one_grad = util.OnlineMeanStd()
reinforce_grad = util.OnlineMeanStd()
reinforce_one_grad = util.OnlineMeanStd()
two_grad = util.OnlineMeanStd()
log_evidence_stats = util.OnlineMeanStd()
log_evidence_grad = util.OnlineMeanStd()
wake_phi_loss_grad = util.OnlineMeanStd()
log_Q_grad = util.OnlineMeanStd()
sleep_loss_grad = util.OnlineMeanStd()
for mc_sample_idx in range(num_mc_samples):
util.print_with_time('MC sample {}'.format(mc_sample_idx))
log_weight, log_q = losses.get_log_weight_and_log_q(
generative_model, inference_network, obss, num_particles)
log_evidence = torch.logsumexp(log_weight, dim=1) - \
np.log(num_particles)
avg_log_evidence = torch.mean(log_evidence)
log_Q = torch.sum(log_q, dim=1)
avg_log_Q = torch.mean(log_Q)
reinforce_one = torch.mean(log_evidence.detach() * log_Q)
reinforce = reinforce_one + avg_log_evidence
vimco_one = 0
for i in range(num_particles):
log_weight_ = log_weight[:, util.range_except(num_particles, i)]
control_variate = torch.logsumexp(
torch.cat([log_weight_, torch.mean(log_weight_, dim=1,
keepdim=True)], dim=1),
dim=1)
vimco_one = vimco_one + (log_evidence.detach() -
control_variate.detach()) * log_q[:, i]
vimco_one = torch.mean(vimco_one)
vimco = vimco_one + avg_log_evidence
normalized_weight = util.exponentiate_and_normalize(log_weight, dim=1)
wake_phi_loss = torch.mean(
-torch.sum(normalized_weight.detach() * log_q, dim=1))
inference_network.zero_grad()
generative_model.zero_grad()
vimco.backward(retain_graph=True)
vimco_grad.update([param.grad for param in
inference_network.parameters()])
inference_network.zero_grad()
generative_model.zero_grad()
vimco_one.backward(retain_graph=True)
vimco_one_grad.update([param.grad for param in
inference_network.parameters()])
inference_network.zero_grad()
generative_model.zero_grad()
reinforce.backward(retain_graph=True)
reinforce_grad.update([param.grad for param in
inference_network.parameters()])
inference_network.zero_grad()
generative_model.zero_grad()
reinforce_one.backward(retain_graph=True)
reinforce_one_grad.update([param.grad for param in
inference_network.parameters()])
inference_network.zero_grad()
generative_model.zero_grad()
avg_log_evidence.backward(retain_graph=True)
two_grad.update([param.grad for param in
inference_network.parameters()])
log_evidence_grad.update([param.grad for param in
generative_model.parameters()])
inference_network.zero_grad()
generative_model.zero_grad()
wake_phi_loss.backward(retain_graph=True)
wake_phi_loss_grad.update([param.grad for param in
inference_network.parameters()])
inference_network.zero_grad()
generative_model.zero_grad()
avg_log_Q.backward(retain_graph=True)
log_Q_grad.update([param.grad for param in
inference_network.parameters()])
log_evidence_stats.update([avg_log_evidence.unsqueeze(0)])
sleep_loss = losses.get_sleep_loss(
generative_model, inference_network, num_particles * len(obss))
inference_network.zero_grad()
generative_model.zero_grad()
sleep_loss.backward()
sleep_loss_grad.update([param.grad for param in
inference_network.parameters()])
return list(map(
lambda x: x.avg_of_means_stds(),
[vimco_grad, vimco_one_grad, reinforce_grad, reinforce_one_grad,
two_grad, log_evidence_stats, log_evidence_grad, wake_phi_loss_grad,
log_Q_grad, sleep_loss_grad]))
def train_rws(
generative_model,
inference_network,
data_loader,
num_iterations,
num_particles,
true_cluster_cov,
test_data_loader,
test_num_particles,
true_generative_model,
checkpoint_path,
):
optimizer_phi = torch.optim.Adam(inference_network.parameters())
optimizer_theta = torch.optim.Adam(generative_model.parameters())
(
theta_losses,
phi_losses,
cluster_cov_distances,
test_log_ps,
test_log_ps_true,
test_kl_qps,
test_kl_pqs,
test_kl_qps_true,
test_kl_pqs_true,
train_log_ps,
train_log_ps_true,
train_kl_qps,
train_kl_pqs,
train_kl_qps_true,
train_kl_pqs_true,
reweighted_train_kl_qps,
reweighted_train_kl_qps_true,
) = ([], [], [], [], [], [], [], [], [], [], [], [], [], [], [], [], [])
data_loader_iter = iter(data_loader)
for iteration in range(num_iterations):
# get obs
try:
obs = next(data_loader_iter)
except StopIteration:
data_loader_iter = iter(data_loader)
obs = next(data_loader_iter)
log_weight, log_q = losses.get_log_weight_and_log_q(
generative_model, inference_network, obs, num_particles)
# wake theta
optimizer_phi.zero_grad()
optimizer_theta.zero_grad()
wake_theta_loss, elbo = losses.get_wake_theta_loss_from_log_weight(
log_weight)
wake_theta_loss.backward(retain_graph=True)
optimizer_theta.step()
# wake phi
optimizer_phi.zero_grad()
optimizer_theta.zero_grad()
wake_phi_loss = losses.get_wake_phi_loss_from_log_weight_and_log_q(
log_weight, log_q)
wake_phi_loss.backward()
optimizer_phi.step()
theta_losses.append(wake_theta_loss.item())
phi_losses.append(wake_phi_loss.item())
cluster_cov_distances.append(
torch.norm(true_cluster_cov -
generative_model.get_cluster_cov()).item())
if iteration % 100 == 0: # test every 100 iterations
(
test_log_p,
test_log_p_true,
test_kl_qp,
test_kl_pq,
test_kl_qp_true,
test_kl_pq_true,
_,
_,
) = models.eval_gen_inf(true_generative_model, generative_model,
inference_network, None, test_data_loader)
test_log_ps.append(test_log_p)
test_log_ps_true.append(test_log_p_true)
test_kl_qps.append(test_kl_qp)
test_kl_pqs.append(test_kl_pq)
test_kl_qps_true.append(test_kl_qp_true)
test_kl_pqs_true.append(test_kl_pq_true)
(
train_log_p,
train_log_p_true,
train_kl_qp,
train_kl_pq,
train_kl_qp_true,
train_kl_pq_true,
_,
_,
reweighted_train_kl_qp,
reweighted_train_kl_qp_true,
) = models.eval_gen_inf(
true_generative_model,
generative_model,
inference_network,
None,
data_loader,
num_particles=num_particles,
reweighted_kl=True,
)
train_log_ps.append(train_log_p)
train_log_ps_true.append(train_log_p_true)
train_kl_qps.append(train_kl_qp)
train_kl_pqs.append(train_kl_pq)
train_kl_qps_true.append(train_kl_qp_true)
train_kl_pqs_true.append(train_kl_pq_true)
reweighted_train_kl_qps.append(reweighted_train_kl_qp)
reweighted_train_kl_qps_true.append(reweighted_train_kl_qp_true)
util.save_checkpoint(
checkpoint_path,
generative_model,
inference_network,
theta_losses,
phi_losses,
cluster_cov_distances,
test_log_ps,
test_log_ps_true,
test_kl_qps,
test_kl_pqs,
test_kl_qps_true,
test_kl_pqs_true,
train_log_ps,
train_log_ps_true,
train_kl_qps,
train_kl_pqs,
train_kl_qps_true,
train_kl_pqs_true,
None,
None,
None,
reweighted_train_kl_qps,
reweighted_train_kl_qps_true,
)
util.print_with_time(
"it. {} | theta loss = {:.2f} | phi loss = {:.2f}".format(
iteration, wake_theta_loss, wake_phi_loss))
# if iteration % 200 == 0:
# z = inference_network.get_latent_dist(obs).sample()
# util.save_plot("images/rws/iteration_{}.png".format(iteration),
# obs[:3], z[:3])
return (
theta_losses,
phi_losses,
cluster_cov_distances,
test_log_ps,
test_log_ps_true,
test_kl_qps,
test_kl_pqs,
test_kl_qps_true,
test_kl_pqs_true,
train_log_ps,
train_log_ps_true,
train_kl_qps,
train_kl_pqs,
train_kl_qps_true,
train_kl_pqs_true,
None,
None,
None,
reweighted_train_kl_qps,
reweighted_train_kl_qps_true,
)
