Skip to content

Recommender [source]

The Recommender module in pnpxai assists you in selecting the most suitable explanation methods for your specific Explainable AI (XAI) needs. It considers factors like modality and neural network architecture.

By analyzing all the available explainers, the recommender module extracts the ones fitting the best to user-specified modality and a model.

Key Features

  • Automated Method Selection: Leverages the provided table to intelligently recommend applicable methods based on your inputs.
  • Comprehensive Coverage: Supports a wide range of explanation methods and metrics for diverse XAI use cases.
  • Easy Integration: Seamlessly integrates into your pnpxai workflows for streamlined explainer development.

Supported Methods and Metrics

The following table summarizes the methods and metrics currently supported by the Recommender module.

Method Data Modalities Architectures
LIME V, L, SD, TS Linear, Convolution, Recurrent, Transformer, Decision Trees
KernelSHAP V, L, SD, TS Linear, Convolution, Recurrent, Transformer, Decision Trees
Gradient V, L, TS Linear, Convolution, Recurrent, Transformer
Gradient × Input V, L, TS Linear, Convolution, Recurrent, Transformer
Grad-CAM V, TS Convolution
Guided Grad-CAM V, TS Convolution
FullGrad V Linear, Convolution, Recurrent, Transformer
SmoothGrad V, L, TS Linear, Convolution, Recurrent, Transformer
VarGrad V, L, TS Linear, Convolution, Recurrent, Transformer
IG V, L, TS Linear, Convolution, Recurrent, Transformer
LRP V, L, TS Linear, Convolution, Recurrent, Transformer
RAP V, L, TS Linear, Convolution, Recurrent, Transformer
AttentionRollout V, L Transformer
TransformerAttribution V, L Transformer

* Supported data modalities are: Vision (V), Language (L), Structured Data (SD), and Time Series (TS)

Usage

  1. Import: Begin by importing the XaiRecommender module in your Python code:

    python from pnpxai.core import XaiRecommender from pnpxai.core.modality import ImageModality

  2. Create an Instance: Construct a XaiRecommender object:

    python model = ... recommender = XaiRecommender() modality = ImageModality()

  3. Get Recommendations: Retrieve recommended methods and metrics by calling XaiRecommender object with your desired parameters:

    python recommender_output = recommender(modality, model) print("Recommended Explainers: ", recommender_output.explainers) print("Detected Architectures: ", recommender_output.detected_architectures)