ENH: implement pooled score feature by FirePheonix · Pull Request #76 · pysal/gwlearn

FirePheonix · 2026-01-25T07:30:08Z

I realized there was a todo with pooled score requirement. This PR implements the same.

# TODO: score_ should be an alias of pooled_score_ - this is different from MGWR

FirePheonix · 2026-01-25T07:31:24Z

If any test cases are to be added for this, or some more changes like other models are to be added with this, then please let me know, i'll do immediately..

codecov · 2026-01-25T07:34:12Z

Codecov Report

❌ Patch coverage is 96.22642% with 2 lines in your changes missing coverage. Please review.
✅ Project coverage is 93.11%. Comparing base (5b8382d) to head (36fb0f2).
⚠️ Report is 4 commits behind head on main.

Files with missing lines	Patch %	Lines
gwlearn/linear_model.py	92.85%	2 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main      #76      +/-   ##
==========================================
+ Coverage   91.36%   93.11%   +1.75%     
==========================================
  Files           6        6              
  Lines         799      872      +73     
==========================================
+ Hits          730      812      +82     
+ Misses         69       60       -9

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

martinfleis

PR titles are used in release notes. Can you have a look how the titles of merged PRs look like and adjust yours, so I don't have to?

martinfleis · 2026-01-25T10:09:16Z

+        if self.oob_y_pooled_.size == 0 or self.oob_pred_pooled_.size == 0:
+            return float("nan")
+        y_true = self.oob_y_pooled_.ravel()
+        y_pred = self.oob_pred_pooled_.ravel()
+        return (y_true == y_pred).mean()


use sklearn.metrics.acccuracy, do not reimplement.

yes, you're right.. changed in latest commit.

martinfleis · 2026-01-25T10:10:21Z

+        if len(self.oob_y_pooled_) == 0:
+            return float("nan")
+        y_true = self.oob_y_pooled_.ravel()
+        y_pred = self.oob_pred_pooled_.ravel()
+        ss_res = ((y_true - y_pred) ** 2).sum()
+        ss_tot = ((y_true - y_true.mean()) ** 2).sum()
+        return 1 - ss_res / ss_tot if ss_tot != 0 else float("nan")


use sklearn.metrics.r2_score, do not reimplement. I don't want to think if this is correct or not. Minimise maintenance burden.

martinfleis · 2026-01-25T10:10:36Z

+            R² computed from all out-of-bag predictions pooled together.
+        """
+        if len(self.oob_y_pooled_) == 0:
+            return float("nan")


Suggested change

return float("nan")

return np.nan

changed in latest commit.

FirePheonix · 2026-01-25T15:39:11Z

may I add tests for this as well? asking because then codecov won't fail. I can make a file named test_pooled_score.py.

martinfleis

I thinking whether this is actually wise or not. We can pool linear models, we can pool OOB values from RF but there's not much we can do about gradient boosting for example. And other models. While we can always leave out focal and alias score_ to focal score. I am not sure what is the best but would rather not rush to get something in before thinking it properly through. What is your take?

martinfleis · 2026-01-26T20:01:09Z

+        # Store pooled y for score computation
+        self.y_pooled_ = y.values
+        self.pred_pooled_ = self.pred_.values


This is wrong. This not not pooled y, this is focal y and focal prediction.

martinfleis · 2026-01-26T20:02:42Z

+        y_true = self.oob_y_pooled_.ravel()
+        y_pred = self.oob_pred_pooled_.ravel()


Do we need ravel() here? How come? This is not flat array?

you're rigth, i'll remove .ravel() from oob_pooled_score and move it to _get_oob_score_data.

FirePheonix · 2026-02-08T03:05:50Z

I thinking whether this is actually wise or not. We can pool linear models, we can pool OOB values from RF but there's not much we can do about gradient boosting for example. And other models. While we can always leave out focal and alias score_ to focal score. I am not sure what is the best but would rather not rush to get something in before thinking it properly through. What is your take?

Yeah, pooling works well for linear models (local y + predictions) and Random Forest (OOB values), but there's no natural mechanism for Gradient Boosting.
So I think we can leave pooled_score_ out for models without proper pooling support rather than aliasing to focal score - less confusing that way.

FirePheonix · 2026-02-25T15:46:51Z

i implemented it for linear models (local y + predictions) and Random Forest, with some tests..
I think, let's let this PR be for now (some other might use this as a reference..), will wait for a proper discussion on this, on what to do about Gradient Boosting and on before moving any forward..

jigyasaba · 2026-02-27T10:32:29Z

Thanks for linking the earlier discussion. I’ll step back from this for now and follow the design discussion first.
Happy to help once the direction is clearer.

FirePheonix · 2026-03-05T19:14:47Z

please lemme know if any updates on this?

martinfleis · 2026-03-05T19:46:42Z

I am simply not convinced that this is the right thing to do, despite the to-do I left in the comment in the code. Both self.y_pooled_ and self.pred_pooled_ are exposed to user so passing them to sklearn function to get accuracy is trivial. The only reason to wrap it would be to have score_ available but again, I am just not sure this is what should be there as it is inconsistent across the models. Would rather have users doing these steps explicitly and knowingly themselves.

pooled-score-feature

d1b52ce

martinfleis reviewed Jan 25, 2026

View reviewed changes

minimize maintainence burden

300d65c

FirePheonix changed the title ~~pooled-score-feature~~ ENH: implement pooled score feature Jan 25, 2026

martinfleis reviewed Jan 26, 2026

View reviewed changes

ravel()-and-focal-y

a326b43

FirePheonix added 5 commits February 8, 2026 08:37

ruff-checks

5bd12c5

Merge remote-tracking branch 'origin/main' into pooled_score_

05356c7

add tests

aafd84f

refined-tests-codecov

0ae38b4

ruff-checks

36fb0f2

martinfleis mentioned this pull request Feb 27, 2026

Added pooled scoring and robust aggregation #97

Closed

		y_true = self.oob_y_pooled_.ravel()
		y_pred = self.oob_pred_pooled_.ravel()

Conversation

FirePheonix commented Jan 25, 2026

Uh oh!

FirePheonix commented Jan 25, 2026

Uh oh!

codecov bot commented Jan 25, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Codecov Report

Uh oh!

martinfleis left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

FirePheonix commented Jan 25, 2026

Uh oh!

martinfleis left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

FirePheonix commented Feb 8, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

FirePheonix commented Feb 25, 2026

Uh oh!

jigyasaba commented Feb 27, 2026

Uh oh!

FirePheonix commented Mar 5, 2026

Uh oh!

martinfleis commented Mar 5, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

codecov bot commented Jan 25, 2026 •

edited

Loading

FirePheonix commented Feb 8, 2026 •

edited

Loading