[LI (ktz03)]

Why are these changes needed?

suggest_config() returns dictionaries from the cached configuration portfolio. Updating a suggested hyperparameter therefore changes later suggestions in the same process. It also removes FLAML_sample_size directly from a caller-provided predictor.

Copy each selected configuration before removing the sample-size key. This keeps the existing selection and returned values while allowing callers to customize a suggestion without changing the predictor or subsequent calls.

For example, using the bundled random-forest regression portfolio:

import numpy as np
from flaml.default import suggest_config

X = np.arange(40).reshape(20, 2)
y = np.arange(20)
config = suggest_config("regression", X, y, "rf", k=1)[0]
original = config["hyperparameters"]["n_estimators"]
config["hyperparameters"]["n_estimators"] = 1
again = suggest_config("regression", X, y, "rf", k=1)[0]
assert again["hyperparameters"]["n_estimators"] == original

The assertion fails before this change: the second call returns the caller's 1 instead of the portfolio's 960.

Related issue number

No existing issue; the reproducer is above.

Checks

• I've used pre-commit to lint the changes in this PR (note the same in integrated in our CI checks).
• I've included any doc changes needed for https://microsoft.github.io/FLAML/. See https://microsoft.github.io/FLAML/docs/Contribute#documentation to build and test documentation locally.
• I've added tests (if relevant) corresponding to the changes introduced in this PR.
• I've made sure all auto checks have passed.

Local validation on Windows with Python 3.12.14, NumPy 1.26.4, pandas 3.0.6 and scikit-learn 1.9.1:

• Nine new regression cases fail on the unchanged base and pass with the fix, covering four bundled estimators for regression/classification and a caller-provided predictor with nested values.
• Four existing related tests pass on both versions: portfolio generation, classification suggestions, Iris AutoML and zero-budget defaults. The combined run is 9 failed/4 passed on the base and 13 passed with the fix; eight unrelated tests are deselected. Both versions retain existing pandas deprecation and estimator convergence warnings.
• Public calls across the eight bundled estimator/task combinations retain the same unmodified suggestions, selection order, k=1 prefix and empty k=0 result.
• pre-commit run --all-files: all 15 checks pass on both the pristine base and the candidate.
• Coverage was measured for flaml/default using the same 13 selected tests; the covered fraction does not decrease. This is focused coverage, not a full-project result.
• The complete optional-dependency test suite and documentation build were not run.

[Copilot]

🟢 Approval recommended

The focused fix preserves selection behavior and includes targeted regression coverage, with no unresolved findings.

No open issues 🎉

What changed in this PR

Prevents callers from modifying FLAML’s cached portfolio or caller-provided predictors through suggested configurations.

Changes:

• Deep-copies selected configurations before removing FLAML_sample_size.
• Documents that returned configurations can be safely modified.
• Adds regression coverage for bundled estimators and nested caller-provided values.

File	Description
test/​default/​test_defaults.py	Adds tests for suggestion independence and predictor preservation.
flaml/​default/​suggest.py	Copies selected configurations and documents mutation safety.

_{Review effort: Balanced}

[Li Jiang (thinkall)]

Overall review of the complete PR: no blocking code issue found; approval withheld pending CI.

Deep-copying each selected portfolio entry before sample-size removal protects both cached and caller-supplied predictors, including nested mutable values, without changing selection order or the result structure. The focused regression command passed 9 tests locally. The review-code harness is clean and Copilot's current Balanced review recommends approval with no open issues. However, the current head has failing build checks, so it does not meet the required all-CI-passed approval gate. This is a formal comment review, not an approval.

Posted by thinkall-agent-auto-reviewer

[Copilot]

🟢 Approval recommended

The focused fix preserves selection behavior and includes targeted regression coverage, with no unresolved findings.

No open issues 🎉

_{Review effort: Balanced}

[Copilot]

🟢 Approval recommended

The localized fix preserves selection behavior and includes targeted regression coverage, with no blocking issues identified.

No open issues 🎉

_{Review effort: Balanced}

[Li Jiang (thinkall)]

Overall review at 3b7f286e1a117bb3eba8fe34125ccfe386b8bb93: re-reviewed the complete PR with review-code, including the copy boundary, predictor cache, caller-owned nested values, and downstream configuration consumers. Copying each selected portfolio entry before cleanup prevents mutations from leaking into later suggestions while preserving selection order and the returned configuration shape. No remaining correctness issue found.

Validation: python -m pytest test\default\test_defaults.py -k 'suggested_config_changes or suggest_config_preserves' -q --disable-warnings passed (9 passed, 12 deselected). The whole-PR Low harness was clean, and its provenance was audited and committed locally. All discussion threads are resolved, current-head Balanced Copilot explicitly recommends approval, check contexts pass, and the actual Build, docs, and Code formatting workflow runs completed successfully on this head.

Posted by thinkall-agent-auto-reviewer