Skip to content
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

Jump to bottom

change LinearBucketEncoder's Linear layer to column-wise #30

Merged
merged 7 commits into from
Aug 24, 2023
Merged

change LinearBucketEncoder's Linear layer to column-wise #30

Changes from 1 commit
Commits
Show all changes
7 commits
Select commit Hold shift + click to select a range
3ce5ced
change LinearBucketEncoder's Linear layer to column-wise
kaidic Aug 24, 2023
4c7223f
update
kaidic Aug 24, 2023
89cffff
update
kaidic Aug 24, 2023
12d7b30
Merge branch 'master' into ple_fix
kaidic Aug 24, 2023
9442d3b
minor
weihua916 Aug 24, 2023
82d4794
minor
weihua916 Aug 24, 2023
b3900db
update
weihua916 Aug 24, 2023
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
21 changes: 11 additions & 10 deletions torch_frame/nn/encoder/stype_encoder.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -59,7 +59,6 @@ def init_modules(self):
for stats in self.stats_list:
num_categories = len(stats[StatType.COUNT][0])
self.embs.append(Embedding(num_categories, self.out_channels))
self.reset_parameters()
kaidic marked this conversation as resolved.
Show resolved Hide resolved

def forward(self, x: Tensor):
r"""Maps input :obj:`x` from TensorFrame (shape [batch_size, num_cols])
Expand Down Expand Up @@ -105,7 +104,6 @@ def init_modules(self):
num_cols = len(self.stats_list)
self.weight = Parameter(torch.empty(num_cols, self.out_channels))
self.bias = Parameter(torch.empty(num_cols, self.out_channels))
self.reset_parameters()
kaidic marked this conversation as resolved.
Show resolved Hide resolved

def forward(self, x: Tensor):
r"""Maps input :obj:`x` from TensorFrame (shape [batch_size, num_cols])
Expand Down Expand Up @@ -151,10 +149,10 @@ def init_modules(self):
quantiles = [stats[StatType.QUANTILES] for stats in self.stats_list]
self.boundaries = torch.tensor(quantiles)
self.interval = self.boundaries[:, 1:] - self.boundaries[:, :-1] + 1e-9

# Adding a linear layer
input_channels = len(self.boundaries[0]) - 1
self.linear_layer = nn.Linear(input_channels, self.out_channels)
kaidic marked this conversation as resolved.
Show resolved Hide resolved
num_cols = len(self.stats_list)
self.weight = Parameter(
torch.empty(num_cols, self.interval.shape[-1], self.out_channels))
self.bias = Parameter(torch.empty(num_cols, self.out_channels))

def forward(self, x: Tensor):
encoded_values = []
Expand All @@ -177,10 +175,13 @@ def forward(self, x: Tensor):
0) + greater_mask * (1 - one_hot_mask)
encoded_values.append(encoded_value)

# Apply linear layer to the encoded values
# Apply column-wise linear transformation
encoded_values = torch.stack(encoded_values, dim=1).squeeze()
return self.linear_layer(encoded_values)
x_lin = torch.einsum('ijk,jkl->ijl', encoded_values, self.weight)
x = x_lin + self.bias
kaidic marked this conversation as resolved.
Show resolved Hide resolved
return x

def reset_parameters(self):
# Reset learnable parameters of the linear layer
self.linear_layer.reset_parameters()
# Reset learnable parameters of the linear transformation
torch.nn.init.normal_(self.weight, std=0.1)
kaidic marked this conversation as resolved.
Show resolved Hide resolved
torch.nn.init.zeros_(self.bias)
Loading