Experimental Modules

TrustLens maintains an experimental ecosystem for cutting-edge research modules that are not yet part of the core production pipeline.

Current Experimental Modules

Explainability / Grad-CAM (trustlens.explainability.gradcam)

  • Purpose: Visual attribution maps for convolutional neural networks.

  • Why Experimental: Requires heavy external dependencies (PyTorch, Torchvision) and is currently limited to vision models.

  • Usage:

    from trustlens.explainability import GradCAM
explainer = GradCAM(model, target_layer)
heatmap = explainer.compute(input_tensor)

Experimental Promotion Criteria

A module is promoted from Experimental to Core when it meets the following standards:

  1. Zero-Crash Stability: Passes internal stress tests without unhandled exceptions.

  2. Standardized API: Implements the standard compute() and get_results_dict() interfaces.

  3. Internal Documentation: Complete docstrings for all public methods.

  4. Unit Test Coverage: Minimum 80% coverage in isolation.

  5. Dependency Management: Dependencies must be optional (extras_require) to keep the core install lightweight.

Contributing to Experimental

We welcome contributions to experimental modules! See Contributing for how to get involved.