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

Add option to save losses at input or patch level #210

Closed
wants to merge 2 commits into from
Closed

Add option to save losses at input or patch level #210

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
2d63cf3
Add option to save losses at input or patch level
brunosan Apr 4, 2024
17a6046
[pre-commit.ci] auto fixes from pre-commit.com hooks
pre-commit-ci[bot] Apr 4, 2024
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
47 changes: 38 additions & 9 deletions src/model_clay.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -792,6 +792,7 @@ def __init__( # noqa: PLR0913
b2=0.95,
embeddings_level: Literal["mean", "patch", "group"] = "mean",
band_groups=None,
save_loss: Literal["False", "Input", "Patch"] = "Patch",
):
super().__init__()
self.save_hyperparameters(logger=True)
Expand Down Expand Up @@ -925,6 +926,25 @@ def predict_step(

assert embeddings_output.shape == torch.Size(expected_size)

# Calculate the loss for the current input if save_loss is "Input" or "Patch"
if self.hparams.save_loss in ["Input", "Patch"]:
loss = self(batch)
else:
loss = None

# Calculate the loss for each patch if save_loss is "Patch"
if self.hparams.save_loss == "Patch":
patch_losses = self.model.per_pixel_loss(
batch["pixels"],
self.model.decoder(
outputs_encoder[0], outputs_encoder[1], outputs_encoder[2]
),
outputs_encoder[3],
)
patch_losses = patch_losses.detach().cpu().numpy()
else:
patch_losses = None

# Create table to store the embeddings with spatiotemporal metadata
unique_epsg_codes = set(int(epsg) for epsg in epsgs)
if len(unique_epsg_codes) == 1: # check that there's only 1 unique EPSG
Expand All @@ -934,16 +954,25 @@ def predict_step(
f"More than 1 EPSG code detected: {unique_epsg_codes}"
)

data = {
"source_url": pd.Series(data=source_urls, dtype="string[pyarrow]"),
"date": pd.to_datetime(arg=dates, format="%Y-%m-%d").astype(
dtype="date32[day][pyarrow]"
),
"embeddings": pa.FixedShapeTensorArray.from_numpy_ndarray(
np.ascontiguousarray(embeddings_output.cpu().detach().__array__())
),
}

if loss is not None:
data["loss"] = loss.item()
if patch_losses is not None:
data["patch_losses"] = pa.FixedSizeListArray.from_arrays(
np.ascontiguousarray(patch_losses), 256
)

gdf = gpd.GeoDataFrame(
data={
"source_url": pd.Series(data=source_urls, dtype="string[pyarrow]"),
"date": pd.to_datetime(arg=dates, format="%Y-%m-%d").astype(
dtype="date32[day][pyarrow]"
),
"embeddings": pa.FixedShapeTensorArray.from_numpy_ndarray(
np.ascontiguousarray(embeddings_output.cpu().detach().__array__())
),
},
data=data,
geometry=shapely.box(
xmin=bboxes[:, 0],
ymin=bboxes[:, 1],
Expand Down
Loading