fishy.analysis.xai

Unified XAI module for deep learning model explanations using LIME and Grad-CAM.

Classes

class fishy.analysis.xai.ExplainerConfig(num_features: int = 5, num_samples: int = 100, output_dir: Path = PosixPath('outputs/xai'), device: str = 'cpu', random_seed: int = 42)

Bases: object

Configuration for model explanation methods.

__init__(num_features: int = 5, num_samples: int = 100, output_dir: Path = PosixPath('outputs/xai'), device: str = 'cpu', random_seed: int = 42) → None

device: str = 'cpu'

num_features: int = 5

num_samples: int = 100

output_dir: Path = PosixPath('outputs/xai')

random_seed: int = 42

class fishy.analysis.xai.GradCAM(model: Module, target_layer: Module)

Bases: object

1D Grad-CAM implementation for analyzing spectral data models.

__init__(model: Module, target_layer: Module) → None

generate_cam(input_tensor: Tensor, target_class: int | None = None)

remove_hooks()

save_activation(module, input, output)

save_gradient(module, grad_input, grad_output)

class fishy.analysis.xai.ModelWrapper(model: Any, device: str)

Bases: object

Wrapper for models to make them compatible with LIME (handles both Torch and Sklearn).

__init__(model: Any, device: str) → None

normalize_intensities(x: ndarray) → ndarray

predict_proba(x: ndarray) → ndarray

Functions

fishy.analysis.xai.run_gradcam_analysis(model, data_loader, device, output_dir, target_layer=None, ctx=None)

fishy.analysis.xai.run_lime_explanation(dataset_name, model, data_module, explainer_config, instance_name, target_label, ctx)

s