for input in tqdm.tqdm(train_input.split(batch_size), desc="vqae-train"):
z = encoder(input)
- zq = z if k < 1 else quantizer(z)
+ zq = z if k < 2 else quantizer(z)
output = decoder(zq)
output = output.reshape(
######################################################################
-# ||x_i - c_j||^2 = ||x_i||^2 + ||c_j||^2 - 2<x_i, c_j>
-def sq2matrix(x, c):
- nx = x.pow(2).sum(1)
- nc = c.pow(2).sum(1)
- return nx[:, None] + nc[None, :] - 2 * x @ c.t()
-
-
-def update_centroids(x, c, nb_min=1):
- _, b = sq2matrix(x, c).min(1)
- b.squeeze_()
- nb_resets = 0
-
- for k in range(0, c.size(0)):
- i = b.eq(k).nonzero(as_tuple=False).squeeze()
- if i.numel() >= nb_min:
- c[k] = x.index_select(0, i).mean(0)
- else:
- n = torch.randint(x.size(0), (1,))
- nb_resets += 1
- c[k] = x[n]
-
- return c, b, nb_resets
-
-
-def kmeans(x, nb_centroids, nb_min=1):
- if x.size(0) < nb_centroids * nb_min:
- print("Not enough points!")
- exit(1)
-
- c = x[torch.randperm(x.size(0))[:nb_centroids]]
- t = torch.full((x.size(0),), -1)
- n = 0
-
- while True:
- c, u, nb_resets = update_centroids(x, c, nb_min)
- n = n + 1
- nb_changes = (u - t).sign().abs().sum() + nb_resets
- t = u
- if nb_changes == 0:
- break
-
- return c, t
-
-
-######################################################################
-
-
def generate_episodes(nb, steps):
all_frames = []
for n in tqdm.tqdm(range(nb), dynamic_ncols=True, desc="world-data"):
train_input = generate_episodes(nb_train_samples, steps)
test_input = generate_episodes(nb_test_samples, steps)
- print(f"{train_input.size()=} {test_input.size()=}")
-
encoder, quantizer, decoder = train_encoder(
train_input, test_input, nb_epochs=nb_epochs
)
pow2 = (2 ** torch.arange(z.size(1), device=z.device))[None, None, :]
z_h, z_w = z.size(2), z.size(3)
- def frame2seq(x):
- z = encoder(x)
- ze_bool = (quantizer(z) >= 0).long()
- seq = (
- ze_bool.permute(0, 2, 3, 1).reshape(ze_bool.size(0), -1, ze_bool.size(1))
- * pow2
- ).sum(-1)
- return seq
-
- def seq2frame(seq, T=1e-2):
- zd_bool = (seq[:, :, None] // pow2) % 2
- zd_bool = zd_bool.reshape(zd_bool.size(0), z_h, z_w, -1).permute(0, 3, 1, 2)
- logits = decoder(zd_bool * 2.0 - 1.0)
- logits = logits.reshape(
- logits.size(0), -1, 3, logits.size(2), logits.size(3)
- ).permute(0, 2, 3, 4, 1)
- results = torch.distributions.categorical.Categorical(
- logits=logits / T
- ).sample()
- return results
+ def frame2seq(input, batch_size=25):
+ seq = []
+
+ for x in input.split(batch_size):
+ z = encoder(x)
+ ze_bool = (quantizer(z) >= 0).long()
+ output = (
+ ze_bool.permute(0, 2, 3, 1).reshape(
+ ze_bool.size(0), -1, ze_bool.size(1)
+ )
+ * pow2
+ ).sum(-1)
+
+ seq.append(output)
+
+ return torch.cat(seq, dim=0)
+
+ def seq2frame(input, batch_size=25, T=1e-2):
+ frames = []
+
+ for seq in input.split(batch_size):
+ zd_bool = (seq[:, :, None] // pow2) % 2
+ zd_bool = zd_bool.reshape(zd_bool.size(0), z_h, z_w, -1).permute(0, 3, 1, 2)
+ logits = decoder(zd_bool * 2.0 - 1.0)
+ logits = logits.reshape(
+ logits.size(0), -1, 3, logits.size(2), logits.size(3)
+ ).permute(0, 2, 3, 4, 1)
+ output = torch.distributions.categorical.Categorical(
+ logits=logits / T
+ ).sample()
+
+ frames.append(output)
+
+ return torch.cat(frames, dim=0)
return train_input, test_input, frame2seq, seq2frame
if __name__ == "__main__":
train_input, test_input, frame2seq, seq2frame = create_data_and_processors(
- 10000, 1000
+ # 10000, 1000,
+ 100,
+ 100,
+ nb_epochs=2,
)
input = test_input[:64]