ISSUES.md

Suggested Issues to open on GitHub

1. Add model export and inference script Title: "Add model export and inference script" Description: Implement src/export.py or examples/inference_example.py that loads trained model and provides predict(wav_path) plus optional Grad-CAM output.

2. Add unit tests for preprocessing Title: "Unit tests for padding and feature extraction" Description: Add tests verifying pad_or_trim, compute_melspec, compute_handcrafted produce expected shapes.

3. Implement TFRecord preprocessing Title: "Precompute mel spectrograms into TFRecord" Description: Add script to precompute features to speed up training.

4. Add TensorBoard logging Title: "Add TensorBoard callback and log directory" Description: Integrate TensorBoard logging in training to track metrics and Grad-CAM images.

5. Add deployment demo (FastAPI) Title: "Create FastAPI demo for inference" Description: Minimal web app to upload WAV and return prediction + Grad-CAM image.

6. Cross-validation experiments Title: "Add k-fold cross-validation experiment" Description: Implement Stratified K-Fold training script and aggregate metrics.

7. Add SHAP result caching & visualization improvements Title: "Improve SHAP plotting: beeswarm, caching" Description: Save SHAP values and add nicer plots for paper.