Dependency cleanup: remove pandas, torchvision, and onnx as core dependencies by coreyjadams · Pull Request #2002 · NVIDIA/physicsnemo

coreyjadams · 2026-09-17T19:46:09Z

PhysicsNeMo Pull Request

Onnx and Torchvision are never actually imported anywhere in the repository. They are removed "for free". Onnx is brought back in the dev list for some tests.

pandas is trickier, we used it in utils for this insolation file. I move that calculation to numpy since it is 100% equivalent to datatime work.

Description

Checklist

I am familiar with the Contributing Guidelines.
New or existing tests cover these changes.
The documentation is up to date with these changes.
The CHANGELOG.md is up to date with these changes.
An issue is linked to this pull request.
If I am implementing a new model or modifying any existing model, I have followed the Models Implementation Coding Standards.

Dependencies

Review Process

All PRs are reviewed by the PhysicsNeMo team before merging.

Depending on which files are changed, GitHub may automatically assign a maintainer for review.

We are also testing AI-based code review tools (e.g., Greptile), which may add automated comments with a confidence score.
This score reflects the AI’s assessment of merge readiness and is not a qualitative judgment of your work, nor is
it an indication that the PR will be accepted / rejected.

AI-generated feedback should be reviewed critically for usefulness.
You are not required to respond to every AI comment, but they are intended to help both authors and reviewers.
Please react to Greptile comments with 👍 or 👎 to provide feedback on their accuracy.

…ndencies

…es as optional dependencies.

copy-pr-bot · 2026-09-17T19:46:12Z

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.

github-actions · 2026-09-17T19:46:20Z

CODEOWNERS review map

Current for commit da1e3ad40ce1. An approval covers every file listed for that owner; one owner is sufficient for shared files.

⏳ @CharlelieLrt — 1 file(s)

physicsnemo/core/version_check.py

⏳ @coreyjadams — 2 file(s)

physicsnemo/core/version_check.py
pyproject.toml

⏳ @ktangsali — 2 file(s)

physicsnemo/core/version_check.py
pyproject.toml

⏳ @mnabian — 1 file(s)

physicsnemo/datapipes/gnn/drivaernet_dataset.py

⏳ @pzharrington — 12 file(s)

physicsnemo/datapipes/healpix/coupledtimeseries_dataset.py
physicsnemo/datapipes/healpix/couplers.py
physicsnemo/datapipes/healpix/timeseries_dataset.py
physicsnemo/experimental/datapipes/healda/configs/sensors.py
physicsnemo/experimental/datapipes/healda/dataset.py
physicsnemo/experimental/datapipes/healda/loaders/era5.py
physicsnemo/experimental/datapipes/healda/loaders/ufs_obs.py
physicsnemo/experimental/datapipes/healda/loaders/zarr_loader.py
physicsnemo/experimental/datapipes/healda/protocols.py
physicsnemo/experimental/datapipes/healda/time_utils.py
physicsnemo/models/dlwp_healpix/HEALPixRecUNet.py
physicsnemo/utils/insolation.py

No CODEOWNER

CHANGELOG.md
uv.lock

Comment /codeowners-info to refresh.

greptile-apps · 2026-09-17T19:53:58Z

Fix the pandas-dependent test collection failure before merging.

Findings

Minimal-install test collection breaks ▶
Coarse-resolution datetime inputs fail ▶

Summary

Removes onnx, torchvision, and pandas from the direct core requirements, moves pandas into feature extras, and converts remaining pandas imports to lazy optional imports.

Replaces pandas-based year-start calculations in insolation with NumPy arithmetic.
Uses string coupler defaults while preserving constructor-time Timedelta normalization.
Updates installation hints, the changelog, and lockfile metadata.
Needs a minimal-install test-collection fix and preservation of coarse-resolution datetime inputs.

_{Reviews (1) · Last reviewed commit: "Remove pandas from utils entirely, pushi..."}

greptile-apps · 2026-09-17T19:54:02Z

    "vtk",
 ]
 datapipes-extras = [
+    "pandas>=2.2.0",


Minimal-install test collection breaks

The minimal-install CI jobs install core plus dev, without model-extras or datapipes-extras, then run all of test/. Moving pandas into these extras leaves that environment without pandas, but test/datapipes/healda/test_time_utils.py:21 still imports it unconditionally. Collection therefore fails with ModuleNotFoundError before the tests can run or skip. Change that import to pytest.importorskip("pandas"), or explicitly include pandas in the test dependencies.

greptile-apps · 2026-09-17T19:54:02Z

-    )
+    # Get the day of year as a float: elapsed days since Jan 1 of each date's year.
+    dates_arr = np.array(dates, dtype="datetime64")
+    days_arr = (dates_arr - dates_arr.astype("datetime64[Y]")) / np.timedelta64(1, "D")


Coarse-resolution datetime inputs fail

For an input such as np.array(["2024-02"], dtype="datetime64[M]"), this subtraction produces timedelta64[M]. Dividing it by timedelta64[D] raises TypeError because NumPy cannot convert variable-length months to days. Previously, the pandas-created January 1 timestamps promoted the subtraction to a fixed-duration unit, so this input worked. Cast the year-start array to datetime64[D] before subtraction, and add regression tests for monthly/yearly inputs alongside sub-day and leap-year dates.

Suggested change

days_arr = (dates_arr - dates_arr.astype("datetime64[Y]")) / np.timedelta64(1, "D")

start_years = dates_arr.astype("datetime64[Y]").astype("datetime64[D]")

days_arr = (dates_arr - start_years) / np.timedelta64(1, "D")

CharlelieLrt

LGTM

mnabian · 2026-09-17T20:10:02Z

-    days_arr = (np.array(dates, dtype="datetime64") - start_years) / np.timedelta64(
-        1, "D"
-    )
+    # Get the day of year as a float: elapsed days since Jan 1 of each date's year.


This implementation is not exactly equivalent. Make sure it does not break downstream tasks.

I added more complexity but Claude thinks it checks out....

mnabian

LGTM

ktangsali

LGTM. Good to check about the downstream applications with the changes to insolation (Mohammad's comment)

ktangsali · 2026-09-17T20:51:45Z

    "treelib>=1.2.5",
    "numpy>=1.22.4",
    "torch>=2.10.0",
-    "torchvision>=0.25.0a0",


Goodbye! You will not be missed 😅

pzharrington · 2026-09-18T02:52:05Z

Insolation was fully untested before?? Sheesh

pzharrington · 2026-09-18T02:53:55Z

+def test_pandas_timestamps_match_datetime():
+    """pandas Timestamps and a tz-aware DatetimeIndex behave like datetime objects."""
+    pd = pytest.importorskip("pandas")
+    ts = [pd.Timestamp("2020-12-31 21:00", tz="US/Eastern")]


Out of pure superstition I think we should test a few more trial dates/timezones (or lack thereof) 😅

pzharrington

LGTM as long as the insolation results are identical (already-released models are trained using that quantity)

coreyjadams added 2 commits September 17, 2026 12:21

Dependency cleanup: remove pandas, torchvision, and onnx as core depe…

6f30732

…ndencies

Remove pandas from utils entirely, pushing it into models and datapip…

da1e3ad

…es as optional dependencies.

coreyjadams requested review from CharlelieLrt, ktangsali, mnabian and pzharrington as code owners September 17, 2026 19:46

greptile-apps Bot reviewed Sep 17, 2026

View reviewed changes

CharlelieLrt approved these changes Sep 17, 2026

View reviewed changes

mnabian reviewed Sep 17, 2026

View reviewed changes

mnabian approved these changes Sep 17, 2026

View reviewed changes

ktangsali approved these changes Sep 17, 2026

View reviewed changes

Align with pandas datetime more exactly

6b440b7

pzharrington reviewed Sep 18, 2026

View reviewed changes

Comment thread test/utils/test_insolation.py

pzharrington Sep 18, 2026

Copy link
Copy Markdown

Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Insolation was fully untested before?? Sheesh

pzharrington reviewed Sep 18, 2026

View reviewed changes

pzharrington approved these changes Sep 18, 2026

View reviewed changes

Merge branch 'main' into refactor/drop-unused-core-deps

f6b984a

	days_arr = (dates_arr - dates_arr.astype("datetime64[Y]")) / np.timedelta64(1, "D")
	start_years = dates_arr.astype("datetime64[Y]").astype("datetime64[D]")
	days_arr = (dates_arr - start_years) / np.timedelta64(1, "D")

Conversation

coreyjadams commented Sep 17, 2026

PhysicsNeMo Pull Request

Description

Checklist

Dependencies

Review Process

Uh oh!

copy-pr-bot Bot commented Sep 17, 2026

Uh oh!

github-actions Bot commented Sep 17, 2026

CODEOWNERS review map

No CODEOWNER

Uh oh!

greptile-apps Bot commented Sep 17, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Findings

Summary

Uh oh!

greptile-apps Bot Sep 17, 2026

Choose a reason for hiding this comment

Uh oh!

greptile-apps Bot Sep 17, 2026

Choose a reason for hiding this comment

Uh oh!

CharlelieLrt left a comment

Choose a reason for hiding this comment

Uh oh!

mnabian Sep 17, 2026

Choose a reason for hiding this comment

Uh oh!

coreyjadams Sep 17, 2026

Choose a reason for hiding this comment

Uh oh!

mnabian left a comment

Choose a reason for hiding this comment

Uh oh!

ktangsali left a comment

Choose a reason for hiding this comment

Uh oh!

ktangsali Sep 17, 2026

Choose a reason for hiding this comment

Uh oh!

pzharrington Sep 18, 2026

Choose a reason for hiding this comment

Uh oh!

pzharrington Sep 18, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

pzharrington left a comment

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants

greptile-apps Bot commented Sep 17, 2026 •

edited

Loading

pzharrington Sep 18, 2026 •

edited

Loading