inspector = ModelInspector(model)

• model: PyTorch model to inspect.

inspect(input_tensor)

• Computes sparsity of convolutional and FC layers.
• Returns (conv_sparsity, fc_sparsity, total_sparsity).

inspect_activations(input_tensor)

• Returns mean, std, and max of activations per layer.

stress_test_distribution_shift(normal_input, shifted_input)

• Returns (normal_mag, shifted_mag) representing mean activation magnitudes.

compute_drift(input_a, input_b)

• Computes activation drift between two inputs.
• Returns a single float drift score.

compute_model_regression(input_tensor, other_inspector)

• Compares activations between two models.
• Returns regression drift score.

compute_failure_attribution(input_tensor, perturbed_tensor)

• Computes layer-wise contribution to output shift.
• Returns dict {layer_name: contribution_score}.

compute_robustness_score(input_tensor, shifted_tensor)

• Aggregates drift, layer sensitivity, and distribution shift into a single score.
• Returns robustness_score.