Feature/lasso-ablation-analysis

[colehanan1]

• Create dedicated ablations/ subfolder within condition output directory
• Move ablation_summary.csv to ablations/ablation_summary.csv
• Move ablation_comparison.png to ablations/ablation_comparison.png
• Individual ablation results remain in separate ablate_* folders

Co-Authored-By: Claude Opus 4.5 noreply@anthropic.com

Summary by CodeRabbit

Release Notes

• New Features

  • Added ablation analysis tool for testing circuit robustness by zeroing receptors and measuring performance impact.
  • Added focus mode analysis tool to evaluate receptor subsets and generate performance curves (MSE vs. number of receptors).

• Documentation

  • Added Robustness Analysis section to README with CLI usage examples and commands.
  • Added comprehensive documentation on stability assessment for LASSO-identified receptor circuits.

_{✏️ Tip: You can customize this high-level summary in your review settings.}

[coderabbitai]

Caution

Review failed

The pull request is closed.

📝 Walkthrough

Walkthrough

Introduces two new CLI scripts for robustness analysis of LASSO-identified receptor circuits: lasso_with_ablations.py (ablation-based testing via receptor zeroing) and lasso_with_focus_mode.py (focus-mode testing on receptor subsets). Includes supporting utilities in the behavioral prediction module, documentation, and comprehensive test coverage.

Changes

Cohort / File(s)	Summary
Documentation README.md, docs/BEHAVIORAL_PREDICTION_ANALYSIS.md	Added "Robustness Analysis" sections describing new CLI scripts, example invocations, and key arguments for ablation and focus mode workflows.
Robustness Analysis Scripts scripts/lasso_with_ablations.py, scripts/lasso_with_focus_mode.py	New CLI entry points implementing end-to-end ablation and focus mode pipelines. Define AblationResult and FocusResult dataclasses, argument parsing, baseline/ablation/focus fitting workflows, CSV/JSON output, and comparison visualizations.
Core Utilities src/door_toolkit/pathways/behavioral_prediction.py	Added five helper functions: resolve_receptor_names (strict name matching), apply_receptor_ablation (zero specified columns), fit_lasso_with_fixed_scaler (LASSO with pre-fitted scaler), restrict_to_receptors (subset feature matrix), and get_top_receptors_by_weight (rank receptors).
Module Exports src/door_toolkit/pathways/__init__.py	Re-exported five new functions from behavioral_prediction module to expand public API surface.
Test Suite tests/test_lasso_ablation.py, tests/test_lasso_focus_mode.py	Comprehensive unit and integration tests covering helper functions, script workflows, file I/O, and edge cases for both ablation and focus mode pipelines.

Sequence Diagram(s)

sequenceDiagram
    actor User
    participant CLI as lasso_with_ablations.py
    participant Predictor as LassoBehavioralPredictor
    participant FileIO as File I/O
    participant Viz as Visualization

    User->>CLI: Run with config + receptors
    CLI->>CLI: Parse args & resolve receptor names
    CLI->>Predictor: Build predictor
    CLI->>Predictor: run_baseline (fit LASSO on full set)
    Predictor-->>CLI: baseline_result (metrics, weights, scaler)
    CLI->>CLI: Ablate receptors (zero columns)
    CLI->>Predictor: fit_lasso_with_fixed_scaler (refit on ablated)
    Predictor-->>CLI: ablation_result (delta metrics)
    CLI->>FileIO: Save per-ablation artifacts
    FileIO-->>CLI: Saved
    CLI->>Viz: plot_ablation_comparison (baseline vs. ablations)
    Viz-->>CLI: PNG generated
    CLI->>FileIO: Save summary CSV
    CLI->>User: Print summary + exit

Loading

sequenceDiagram
    actor User
    participant CLI as lasso_with_focus_mode.py
    participant Predictor as LassoBehavioralPredictor
    participant Ranker as Receptor Ranking
    participant FileIO as File I/O
    participant Viz as Visualization

    User->>CLI: Run with config + focus targets
    CLI->>CLI: Parse args
    CLI->>Predictor: Build predictor
    CLI->>Predictor: run_baseline (fit LASSO on full set)
    Predictor-->>CLI: baseline (weights, R², MSE, scaler)
    CLI->>Ranker: get_top_receptors_by_weight (rank by |weight|)
    Ranker-->>CLI: Ranked receptor list
    CLI->>CLI: Generate focus sets (explicit or top-N)
    loop For each focus set
        CLI->>CLI: restrict_to_receptors (subset features)
        CLI->>Predictor: fit_lasso_with_fixed_scaler (refit on subset)
        Predictor-->>CLI: focus_result (subset metrics)
    end
    CLI->>FileIO: Save baseline.json, focus JSONs, focus_curve.csv
    FileIO-->>CLI: Saved
    CLI->>Viz: plot_focus_curve (MSE & R² vs. N receptors)
    Viz-->>CLI: PNG generated
    CLI->>User: Print summary + exit

Loading

Estimated code review effort

🎯 4 (Complex) | ⏱️ ~60 minutes

Poem

🐰 Two scripts hop into the fray,
Ablations and focus to test the way,
Receptors ablated, weights ranked high,
Robustness measured with a keen eye!
Behold the curves and the summary stats,
Code review, pass — no ifs or brats! 🌿

✨ Finishing touches

• 📝 Generate docstrings

---

📜 Recent review details

Configuration used: defaults

Review profile: CHILL

Plan: Pro

📥 Commits

Reviewing files that changed from the base of the PR and between bc9234f and 1ed3db4.

📒 Files selected for processing (8)

• README.md
• docs/BEHAVIORAL_PREDICTION_ANALYSIS.md
• scripts/lasso_with_ablations.py
• scripts/lasso_with_focus_mode.py
• src/door_toolkit/pathways/__init__.py
• src/door_toolkit/pathways/behavioral_prediction.py
• tests/test_lasso_ablation.py
• tests/test_lasso_focus_mode.py

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

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