Skip to content

Fix gumbel #4

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
wants to merge 2 commits into
base: main
Choose a base branch
from
Open

Fix gumbel #4

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
ee3ca19
fix gumbel-softmax sampling
alexander-telepov Apr 18, 2022
80f1edb
fix log_prob padding
alexander-telepov Apr 18, 2022
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
56 changes: 26 additions & 30 deletions core_motif.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -267,7 +267,7 @@ def create_candidate_motifs(self):


def gumbel_softmax(self, logits: torch.Tensor, tau: float = 1, hard: bool = False, eps: float = 1e-10, dim: int = -1, \
g_ratio: float = 1e-3) -> torch.Tensor:
g_ratio: float = 1.) -> torch.Tensor:
gumbels = (
-torch.empty_like(logits, memory_format=torch.legacy_contiguous_format).exponential_().log()
) # ~Gumbel(0,1)
Expand Down Expand Up @@ -328,18 +328,17 @@ def forward(self, graph_emb, node_emb, g, cands, deterministic=False):
if g.batch_size != 1:
ac_first_prob = [torch.softmax(logit, dim=0)
for i, logit in enumerate(torch.split(logits_first, att_len, dim=0))]
ac_first_prob = [p+1e-8 for p in ac_first_prob]
log_ac_first_prob = [x.log() for x in ac_first_prob]

else:
ac_first_prob = torch.softmax(logits_first, dim=0) + 1e-8
ac_first_prob = torch.softmax(logits_first, dim=0)
log_ac_first_prob = ac_first_prob.log()

if g.batch_size != 1:
first_stack = []
first_ac_stack = []
for i, node_emb_i in enumerate(torch.split(att_emb, att_len, dim=0)):
ac_first_hot_i = self.gumbel_softmax(ac_first_prob[i], tau=self.tau, hard=True, dim=0).transpose(0,1)
ac_first_hot_i = self.gumbel_softmax(log_ac_first_prob[i], tau=self.tau, hard=True, dim=0).transpose(0,1)
ac_first_i = torch.argmax(ac_first_hot_i, dim=-1)
first_stack.append(torch.matmul(ac_first_hot_i, node_emb_i))
first_ac_stack.append(ac_first_i)
Expand All @@ -355,20 +354,20 @@ def forward(self, graph_emb, node_emb, g, cands, deterministic=False):
for i, ac_first_prob_i in enumerate(ac_first_prob)], dim=0).contiguous()

log_ac_first_prob = torch.cat([
torch.cat([log_ac_first_prob_i, log_ac_first_prob_i.new_zeros(
max(self.max_action - log_ac_first_prob_i.size(0),0),1)]
torch.cat([log_ac_first_prob_i, torch.full(
(max(self.max_action - log_ac_first_prob_i.size(0),0),1), float("-inf"), device=self.device)]
, 0).contiguous().view(1,self.max_action)
for i, log_ac_first_prob_i in enumerate(log_ac_first_prob)], dim=0).contiguous()

else:
ac_first_hot = self.gumbel_softmax(ac_first_prob, tau=self.tau, hard=True, dim=0).transpose(0,1)
ac_first_hot = self.gumbel_softmax(log_ac_first_prob, tau=self.tau, hard=True, dim=0).transpose(0,1)
ac_first = torch.argmax(ac_first_hot, dim=-1)
emb_first = torch.matmul(ac_first_hot, att_emb)
ac_first_prob = torch.cat([ac_first_prob, ac_first_prob.new_zeros(
max(self.max_action - ac_first_prob.size(0),0),1)]
, 0).contiguous().view(1,self.max_action)
log_ac_first_prob = torch.cat([log_ac_first_prob, log_ac_first_prob.new_zeros(
max(self.max_action - log_ac_first_prob.size(0),0),1)]
log_ac_first_prob = torch.cat([log_ac_first_prob, torch.full(
(max(self.max_action - log_ac_first_prob.size(0),0),1), float("-inf"), device=self.device)]
, 0).contiguous().view(1,self.max_action)

# ===============================
Expand All @@ -381,18 +380,16 @@ def forward(self, graph_emb, node_emb, g, cands, deterministic=False):
self.action2_layers[1](cand_expand) + self.action2_layers[2](emb_first_expand)

logit_second = self.action2_layers[3](emb_cat).squeeze(-1)
ac_second_prob = F.softmax(logit_second, dim=-1) + 1e-8
ac_second_prob = F.softmax(logit_second, dim=-1)
log_ac_second_prob = ac_second_prob.log()

ac_second_hot = self.gumbel_softmax(ac_second_prob, tau=self.tau, hard=True, g_ratio=1e-3)
ac_second_hot = self.gumbel_softmax(log_ac_second_prob, tau=self.tau, hard=True)
emb_second = torch.matmul(ac_second_hot, cand_graph_emb)
ac_second = torch.argmax(ac_second_hot, dim=-1)

# Print gumbel otuput
ac_second_gumbel = self.gumbel_softmax(ac_second_prob, tau=self.tau, hard=False, g_ratio=1e-3)


# ===============================
# step 4 : where to add on motif
# step 3 : where to add on motif
# ===============================
# Select att points from candidate
cand_att_emb = torch.masked_select(cand_node_emb, cand_att_mask.unsqueeze(-1))
Expand All @@ -418,20 +415,19 @@ def forward(self, graph_emb, node_emb, g, cands, deterministic=False):
if g.batch_size != 1:
ac_third_prob = [torch.softmax(logit,dim=-1)
for i, logit in enumerate(torch.split(logits_third.squeeze(-1), ac3_att_len, dim=0))]
ac_third_prob = [p+1e-8 for p in ac_third_prob]
log_ac_third_prob = [x.log() for x in ac_third_prob]

else:
logits_third = logits_third.transpose(1,0)
ac_third_prob = torch.softmax(logits_third, dim=-1) + 1e-8
ac_third_prob = torch.softmax(logits_third, dim=-1)
log_ac_third_prob = ac_third_prob.log()

# gumbel softmax sampling and zero-padding
if g.batch_size != 1:
third_stack = []
third_ac_stack = []
for i, node_emb_i in enumerate(torch.split(emb_cat_ac3, ac3_att_len, dim=0)):
ac_third_hot_i = self.gumbel_softmax(ac_third_prob[i], tau=self.tau, hard=True, dim=-1)
ac_third_hot_i = self.gumbel_softmax(log_ac_third_prob[i], tau=self.tau, hard=True, dim=-1)
ac_third_i = torch.argmax(ac_third_hot_i, dim=-1)
third_stack.append(torch.matmul(ac_third_hot_i, node_emb_i))
third_ac_stack.append(ac_third_i)
Expand All @@ -446,21 +442,21 @@ def forward(self, graph_emb, node_emb, g, cands, deterministic=False):
for i, ac_third_prob_i in enumerate(ac_third_prob)], dim=0).contiguous()

log_ac_third_prob = torch.cat([
torch.cat([log_ac_third_prob_i, log_ac_third_prob_i.new_zeros(
self.max_action - log_ac_third_prob_i.size(0))]
torch.cat([log_ac_third_prob_i, torch.full(
(self.max_action - log_ac_third_prob_i.size(0), ), float("-inf"), device=self.device)]
, 0).contiguous().view(1,self.max_action)
for i, log_ac_third_prob_i in enumerate(log_ac_third_prob)], dim=0).contiguous()

else:
ac_third_hot = self.gumbel_softmax(ac_third_prob, tau=self.tau, hard=True, dim=-1)
ac_third_hot = self.gumbel_softmax(log_ac_third_prob, tau=self.tau, hard=True, dim=-1)
ac_third = torch.argmax(ac_third_hot, dim=-1)
emb_third = torch.matmul(ac_third_hot, emb_cat_ac3)

ac_third_prob = torch.cat([ac_third_prob, ac_third_prob.new_zeros(
1, self.max_action - ac_third_prob.size(1))]
, -1).contiguous()
log_ac_third_prob = torch.cat([log_ac_third_prob, log_ac_third_prob.new_zeros(
1, self.max_action - log_ac_third_prob.size(1))]
log_ac_third_prob = torch.cat([log_ac_third_prob, torch.full(
(1, self.max_action - log_ac_third_prob.size(1)), float("-inf"), device=self.device)]
, -1).contiguous()

# ==== concat everything ====
Expand All @@ -470,7 +466,7 @@ def forward(self, graph_emb, node_emb, g, cands, deterministic=False):
log_ac_second_prob, log_ac_third_prob], dim=1).contiguous()
ac = torch.stack([ac_first, ac_second, ac_third], dim=1)

return ac, (ac_prob, log_ac_prob), (ac_first_prob, ac_second_hot, ac_third_prob)
return ac, (ac_prob, (log_ac_prob.exp() + 1e-8).log()), (ac_first_prob, ac_second_hot, ac_third_prob)

def sample(self, ac, graph_emb, node_emb, g, cands):
g.ndata['node_emb'] = node_emb
Expand Down Expand Up @@ -504,8 +500,8 @@ def sample(self, ac, graph_emb, node_emb, g, cands):
max(self.max_action - ac_first_prob.size(1),0))]
, 1).contiguous()

log_ac_first_prob = torch.cat([log_ac_first_prob, log_ac_first_prob.new_zeros(1,
max(self.max_action - log_ac_first_prob.size(1),0))]
log_ac_first_prob = torch.cat([log_ac_first_prob, torch.full((1,
max(self.max_action - log_ac_first_prob.size(1),0)), float("-inf"), device=self.device)]
, 1).contiguous()
emb_first = att_emb[ac[0]].unsqueeze(0)

Expand All @@ -522,7 +518,7 @@ def sample(self, ac, graph_emb, node_emb, g, cands):
ac_second_prob = F.softmax(logit_second, dim=-1) + 1e-8
log_ac_second_prob = ac_second_prob.log()

ac_second_hot = self.gumbel_softmax(ac_second_prob, tau=self.tau, hard=True, g_ratio=1e-3)
ac_second_hot = self.gumbel_softmax(logit_second, tau=self.tau, hard=True)
emb_second = torch.matmul(ac_second_hot, cand_graph_emb)
ac_second = torch.argmax(ac_second_hot, dim=-1)

Expand Down Expand Up @@ -559,16 +555,16 @@ def sample(self, ac, graph_emb, node_emb, g, cands):
ac_third_prob = torch.cat([ac_third_prob, ac_third_prob.new_zeros(
1, self.max_action - ac_third_prob.size(1))]
, -1).contiguous()
log_ac_third_prob = torch.cat([log_ac_third_prob, log_ac_third_prob.new_zeros(
1, self.max_action - log_ac_third_prob.size(1))]
log_ac_third_prob = torch.cat([log_ac_third_prob, torch.full(
(1, self.max_action - log_ac_third_prob.size(1)), float("-inf"), device=self.device)]
, -1).contiguous()

# ==== concat everything ====
ac_prob = torch.cat([ac_first_prob, ac_second_prob, ac_third_prob], dim=1).contiguous()
log_ac_prob = torch.cat([log_ac_first_prob,
log_ac_second_prob, log_ac_third_prob], dim=1).contiguous()

return (ac_prob, log_ac_prob), (ac_first_prob, ac_second_hot, ac_third_prob)
return (ac_prob, (log_ac_prob.exp() + 1e-8).log()), (ac_first_prob, ac_second_hot, ac_third_prob)


class GCNEmbed(nn.Module):
Expand Down
Loading