[Feature] Train fix success predictor

[ionfwsrijan]

Description

Train a logistic regression model that predicts whether a proposed fix will pass verification, given its features.

What to implement

Create backend/scripts/train_fix_predictor.py:

• Reads CSV from Issue 5.2 (or directly from SQLite)
• Encodes categorical features, trains LogisticRegression
• Reports accuracy and ROC-AUC on 20% held-out split
• Saves to backend/app/ml/models/fix_predictor.pkl
• Requires ≥ 100 examples; exits with message if fewer

Acceptance criteria

• Produces fix_predictor.pkl
• Prints ROC-AUC score to stdout
• fix_predictor.pkl is gitignored
• Script has --help flag

[diksha78dev]

@ionfwsrijan I would like to work on under the SSoC'26 programs.

Proposed Solution

To implement the fix success predictor, I will develop a Python script that leverages Scikit-Learn to process data and train the model.

• Action: Create backend/scripts/train_fix_predictor.py.
• Implementation:
  • Data Loading: Use pandas to read the training data from the relevant CSV or SQLite database source.
  • Preprocessing: Implement categorical feature encoding (e.g., using OneHotEncoder or LabelEncoder) and data normalization.
  • Model Training: Use sklearn.linear_model.LogisticRegression to train the predictor.
  • Evaluation: Perform a 80/20 train-test split, calculate accuracy and ROC-AUC scores, and print these metrics to stdout.
  • Validation: Add a check to verify the dataset contains $\ge 100$ examples, exiting with a descriptive error message if insufficient.
  • Persistence: Save the trained model object as a .pkl file in backend/app/ml/models/fix_predictor.pkl.
  • CLI UX: Add a --help flag using argparse to document script usage.
• Maintenance: Ensure fix_predictor.pkl is added to .gitignore.

[ionfwsrijan]

@diksha78dev Your implementation plan is great but as this issue depend on issue #178 and #179, i'd request you to ask for assignment on #178.

[diksha78dev]

Ok

[prasiddhi-105]

Hii please assign this issue to me under SSOC'26!!!

[github-actions]

⚠️ Assignment Notice

Hi @prasiddhi-105! 👋

It looks like you currently have 2 open assigned issues.

To ensure fair distribution of work and timely progress, we encourage contributors to complete or unassign their existing issues before taking on additional ones.

📋 Current Assigned Issues

• Deduplication executes on empty embeddings due to Pydantic/dict type mismatch - all deduplication meaningless #259 — Deduplication executes on empty embeddings due to Pydantic/dict type mismatch - all deduplication meaningless
• [Feature] Train fix success predictor #180 — [Feature] Train fix success predictor

If you're actively working on them, feel free to continue. Otherwise, please consider freeing up an issue for other contributors.

Note: This is only a friendly reminder. Your assignment has not been removed.

Thank you for contributing to PatchPilot! 🚀

[ionfwsrijan]

@prasiddhi-105 Issue assigned. Please star the repo