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

Fix/dlinear padding for even kernel size #1695

Merged
merged 4 commits into from
Apr 11, 2023
Merged

Fix/dlinear padding for even kernel size #1695

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
4ce2620
fix: asymetrical padding when using even kernel size, moved the compu…
madtoinou Apr 5, 2023
1f55719
feat: added unitest to cover both even and odd kernel size values
madtoinou Apr 5, 2023
1cf8f98
Merge branch 'master' into fix/dlinear_padding
madtoinou Apr 10, 2023
840396b
Merge branch 'master' into fix/dlinear_padding
dennisbader Apr 11, 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
17 changes: 12 additions & 5 deletions darts/models/forecasting/dlinear.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -24,13 +24,19 @@ class _MovingAvg(nn.Module):

def __init__(self, kernel_size, stride):
super().__init__()
self.kernel_size = kernel_size
# asymmetrical padding, shorther on the ts start side
if kernel_size % 2 == 0:
self.padding_size_left = kernel_size // 2 - 1
self.padding_size_right = kernel_size // 2
else:
self.padding_size_left = (kernel_size - 1) // 2
self.padding_size_right = (kernel_size - 1) // 2
self.avg = nn.AvgPool1d(kernel_size=kernel_size, stride=stride, padding=0)

def forward(self, x):
# padding on the both ends of time series
front = x[:, 0:1, :].repeat(1, (self.kernel_size - 1) // 2, 1)
end = x[:, -1:, :].repeat(1, (self.kernel_size - 1) // 2, 1)
# padding on the both ends of time series with the extremities values
front = x[:, 0:1, :].repeat(1, self.padding_size_left, 1)
end = x[:, -1:, :].repeat(1, self.padding_size_right, 1)
x = torch.cat([front, x, end], dim=1)
x = self.avg(x.permute(0, 2, 1))
x = x.permute(0, 2, 1)
Expand Down Expand Up @@ -254,7 +260,8 @@ def __init__(
Default: False.

kernel_size
The size of the kernel for the moving average (default=25).
The size of the kernel for the moving average (default=25). If the size of the kernel is even,
the padding will be asymmetrical (shorter on the start/left side).
const_init
Whether to initialize the weights to 1/in_len. If False, the default PyTorch
initialization is used (default='True').
Expand Down
7 changes: 6 additions & 1 deletion darts/tests/models/forecasting/test_dlinear_nlinear.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -59,13 +59,18 @@ def test_fit(self):
large_ts = tg.constant_timeseries(length=100, value=1000)
small_ts = tg.constant_timeseries(length=100, value=10)

for model_cls in [DLinearModel, NLinearModel]:
for (model_cls, kwargs) in [
(DLinearModel, {"kernel_size": 5}),
(DLinearModel, {"kernel_size": 6}),
(NLinearModel, {}),
]:
# Test basic fit and predict
model = model_cls(
input_chunk_length=1,
output_chunk_length=1,
n_epochs=10,
random_state=42,
**kwargs
)
model.fit(large_ts[:98])
pred = model.predict(n=2).values()[0]
Expand Down