mixture = VariationalMixture(
xdim=xdim,
hdim=hdim,
n_hidden=n_hidden,
n_classes=n_classes,
components=[
RealNVP(n_blocks=n_realnvp_blocks,
xdim=xdim,
hdim=hdim_realnvp,
n_hidden=n_hidden_realnvp,
base_dist=base_dist) for _ in range(n_classes)
],
)
# %%
count_parameters(mixture)
# %%
x = np.linspace(-1, 1, 1000)
z = np.array(np.meshgrid(x, x)).transpose(1, 2, 0)
z = np.reshape(z, [z.shape[0] * z.shape[1], -1])
mesh = z.reshape([1000, 1000, 2]).transpose(2, 0, 1)
xx = mesh[0]
yy = mesh[1]
def _pre_backward_callback(mixture, n_iter, log_probs, prior_crossent,
q_entropy, temperature, writer):
if n_iter % 5 == 0:
writer.add_scalar('train/log_probs', log_probs, n_iter)
colors[idx_3] = 3
# %%
plt.scatter(X0[:, 0], X0[:, 1], s=5, c=colors)
# %%
flow = RealNVP(n_blocks=4,
input_size=2,
hidden_size=32,
n_hidden=1,
base_dist=base_dist)
opt = optim.Adam(flow.parameters(), lr=1e-2)
# %%
count_parameters(flow)
# %%
n_epochs = 500
bs = 256
# %%
writer = SummaryWriter(f"../tensorboard_logs/{now_str()}")
dataloader = DataLoader(X, batch_size=bs, shuffle=True, num_workers=0)
best_loss = torch.Tensor([float("+inf")])
#attempts = 0
for epoch in trange(n_epochs):
batches = range((len(X) - 1) // bs + 1)
for i, xb in enumerate(dataloader):
c = torch.Tensor(c)
plt.figure(figsize=(10, 10))
plt.scatter(X[:, 0].numpy(), X[:, 1].numpy(), c=c.numpy(), s=5)
plt.show()
# %%
xdim = 2
hdim = 2
n_hidden = 3
n_classes = 3
mlp = MLP(xdim=xdim, hdim=hdim, n_hidden=n_hidden, n_classes=n_classes)
# %%
count_parameters(mlp)
# %%
n_epochs = 10000
bs = 512
opt = optim.Adam(mlp.parameters(), lr=0.001)
# %%
c_onehot = torch_onehot(c, 3)
# %%
writer = SummaryWriter(f"/workspace/sandbox/tensorboard_logs/{now_str()}")
best_loss = torch.Tensor([float("+inf")])
attempts = 0