Use new input validation infrastructure for cuml.decomposition.

[csadorf]

This applies the new input validation utilities added in #7973 to cuml.decomposition.

Doing this fixed ~9 failing sklearn compatibility tests for cuml proper, and at least 36 upstream tests for cuml.accel.

Fixes #7990
Part of #7428

[coderabbitai]

Note

Reviews paused

It looks like this branch is under active development. To avoid overwhelming you with review comments due to an influx of new commits, CodeRabbit has automatically paused this review. You can configure this behavior by changing the reviews.auto_review.auto_pause_after_reviewed_commits setting.

Use the following commands to manage reviews:

• @coderabbitai resume to resume automatic reviews.
• @coderabbitai review to trigger a single review.

Use the checkboxes below for quick actions:

• ▶️ Resume reviews
• 🔍 Trigger review

📝 Walkthrough

Walkthrough

Replace bespoke array conversion and sparse helpers in IncrementalPCA, PCA, and TruncatedSVD with unified validation (check_inputs/check_array), allocate buffers with CuPy and wrap as CumlArray, update reflect decorators to reset="type", and adjust tests/xfails to expect passing sklearn checks.

Changes

Cohort / File(s)	Summary
IncrementalPCA python/cuml/cuml/decomposition/incremental_pca.py	Unifies validation via check_inputs/check_array (accepts csr/csc, float32/64), removes _validate_sparse_input, densifies sparse batches before partial_fit, changes partial_fit sparse validation flow, and updates fit decorator to @cuml.internals.reflect(reset="type").
PCA (Cython) python/cuml/cuml/decomposition/pca.pyx	Switches to check_inputs/check_array (dtype + order="F"), allocates intermediates with cupy.zeros(..., order="F"), uses .data.ptr for kernels, wraps outputs as CumlArray(data=...), updates dense helpers to accept index=None, replaces output-type context manager with explicit .to_output("cupy"), and changes reflect decorators to reset="type".
TruncatedSVD (Cython) python/cuml/cuml/decomposition/tsvd.pyx	Replaces input_to_cuml_array with check_inputs/check_array (captures index), allocates outputs with cupy.zeros(..., order="F") and constructs CumlArray(data=..., index=...), uses .data.ptr for kernels, removes check_features from transform, adds inverse-transform column-count guard, and updates reflect decorator to reset="type".
Tests & xfail config python/cuml/cuml_accel_tests/upstream/scikit-learn/xfail-list.yaml, python/cuml/tests/test_incremental_pca.py, python/cuml/tests/test_sklearn_compatibility.py, python/cuml/tests/test_exceptions.py	Removes several xfail entries and conditional skips for PCA/IncrementalPCA/TruncatedSVD (dtype/object, empty-data messages, NaN/Inf, transform checks), enables csc sparse cases in IncrementalPCA tests, and relaxes a sparse-related regex in an exception test.

Estimated code review effort

🎯 4 (Complex) | ⏱️ ~45 minutes

Possibly related PRs

• Forward-merge release/26.04 into main #7894: Related changes to reflect/reset behavior and decorator updates (use of reset="type").
• CI Update xfail list #7768: Related adjustments to sklearn-compatibility xfail mappings for PCA/TruncatedSVD/IncrementalPCA.

Suggested reviewers

• dantegd
• jcrist
• viclafargue

🚥 Pre-merge checks | ✅ 4 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 57.14% which is insufficient. The required threshold is 80.00%.	Write docstrings for the functions missing them to satisfy the coverage threshold.

✅ Passed checks (4 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title 'Use new input validation infrastructure for cuml.decomposition' directly and concisely describes the primary change—applying new validation utilities to the decomposition module.
Description check	✅ Passed	The description accurately describes the changeset: applying new input validation utilities to cuml.decomposition, references the foundational PR #7973, and mentions both the test fixes and linked issues.
Linked Issues check	✅ Passed	The code changes fully address issue #7990 by migrating PCA, IncrementalPCA, and TruncatedSVD to use the new check_inputs/check_array validation utilities, completing the decomposition module's transition to standardized input validation.
Out of Scope Changes check	✅ Passed	All changes are directly scoped to the decomposition module's validation infrastructure migration; test updates and xfail removals are necessary consequences of the validation improvements and not extraneous additions.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests

Warning

Review ran into problems

🔥 Problems

Timed out fetching pipeline failures after 30000ms

---

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

[coderabbitai]

Actionable comments posted: 2

🤖 Prompt for all review comments with AI agents

Verify each finding against the current code and only fix it if needed.

Inline comments:
In `@python/cuml/cuml/decomposition/incremental_pca.py`:
- Around line 277-283: The code resets input metadata unconditionally using
n_samples_seen_, _set_output_type and check_features even when check_input is
False; wrap the initial metadata setup so it only runs if check_input is True:
compute first_call and call self._set_output_type(X) and check_features(self, X,
reset=first_call) inside an if check_input: block (or ensure check_features is
invoked with reset=False when check_input is False). Update the partial_fit/fit
flow to preserve the original reflected input type and feature_names_in_ by
avoiding these resets for internal CuPy batches.
- Around line 222-228: In fit() the call to check_array strips pandas/cuDF
feature names before check_features(..., reset=True) runs, so feature_names_in_
and n_features_in_ never get set; fix by capturing feature names and resetting
feature metadata before coercion—either call check_features(self, X, reset=True)
before check_array or switch to check_inputs(X, accept_sparse=..., dtype=...,
convert_dtype=..., reset=True) so names are preserved, and ensure fit() assigns
n_features_in_ and feature_names_in_ (the learned attributes) after validation;
update references to check_array, check_features, check_inputs,
feature_names_in_, and n_features_in_ accordingly.

🪄 Autofix (Beta)

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro Plus

Run ID: f36de441-99a0-4085-8c87-f74bfad41870

📥 Commits

Reviewing files that changed from the base of the PR and between c7b104c and a2b019b.

📒 Files selected for processing (6)

• python/cuml/cuml/decomposition/incremental_pca.py
• python/cuml/cuml/decomposition/pca.pyx
• python/cuml/cuml/decomposition/tsvd.pyx
• python/cuml/cuml_accel_tests/upstream/scikit-learn/xfail-list.yaml
• python/cuml/tests/test_incremental_pca.py
• python/cuml/tests/test_sklearn_compatibility.py

💤 Files with no reviewable changes (3)

• python/cuml/tests/test_incremental_pca.py
• python/cuml/tests/test_sklearn_compatibility.py
• python/cuml/cuml_accel_tests/upstream/scikit-learn/xfail-list.yaml

[coderabbitai]

Caution

Some comments are outside the diff and can’t be posted inline due to platform limitations.

⚠️ Outside diff range comments (1)

python/cuml/cuml/decomposition/incremental_pca.py (1)

431-439: ⚠️ Potential issue | 🟡 Minor

Potential AttributeError when transform is called on sparse data after partial_fit-only usage.

batch_size_ is set only in fit() (line 236-238), not in partial_fit(). If a user trains exclusively via partial_fit() calls and then calls transform() on sparse input, line 435 will raise AttributeError: 'IncrementalPCA' object has no attribute 'batch_size_'.

Consider falling back to a default batch size when batch_size_ is not set:

Suggested fix

+            batch_size = getattr(self, "batch_size_", None)
+            if batch_size is None:
+                batch_size = 5 * X.shape[1]
             for batch in _gen_batches(
                 n_samples,
-                self.batch_size_,
+                batch_size,
                 min_batch_size=self.n_components or 0,
             ):

🤖 Prompt for AI Agents

Verify each finding against the current code and only fix it if needed.

In `@python/cuml/cuml/decomposition/incremental_pca.py` around lines 431 - 439,
transform()'s sparse-path uses self.batch_size_ which is only set in fit(),
causing AttributeError when the user trained only via partial_fit(); update
transform (sparse branch) to fall back to a default batch size when batch_size_
is missing (e.g., use getattr(self, "batch_size_", some_default) or compute a
safe default from n_samples/n_components) before calling _gen_batches so
_transform_sparse and _gen_batches receive a valid batch range; touch the
transform method and references to batch_size_ (and possibly __init__ or
partial_fit) to ensure consistency with partial_fit-only workflows and preserve
min_batch_size=self.n_components or 0 behavior.

🤖 Prompt for all review comments with AI agents

Verify each finding against the current code and only fix it if needed.

Outside diff comments:
In `@python/cuml/cuml/decomposition/incremental_pca.py`:
- Around line 431-439: transform()'s sparse-path uses self.batch_size_ which is
only set in fit(), causing AttributeError when the user trained only via
partial_fit(); update transform (sparse branch) to fall back to a default batch
size when batch_size_ is missing (e.g., use getattr(self, "batch_size_",
some_default) or compute a safe default from n_samples/n_components) before
calling _gen_batches so _transform_sparse and _gen_batches receive a valid batch
range; touch the transform method and references to batch_size_ (and possibly
__init__ or partial_fit) to ensure consistency with partial_fit-only workflows
and preserve min_batch_size=self.n_components or 0 behavior.

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro Plus

Run ID: 9fccc59e-4423-4549-aaec-8d4fd2e32323

📥 Commits

Reviewing files that changed from the base of the PR and between 527798c and 3063ec6.

📒 Files selected for processing (2)

• python/cuml/cuml/decomposition/incremental_pca.py
• python/cuml/tests/test_exceptions.py

[jcrist]

Looks good, thanks! Looks like a small merge conflict, but otherwise !

[jcrist]

This check_input kwarg is odd, but it's what sklearn does too so 🤷. I do think there's a way to make this work with check_inputs instead of check_features + check_array, but what you have here is fine too.

[csadorf]

/merge