How kush-kodes-4/xaio_iit works
This repository contains XAIO, a model-agnostic local-explanation method, plus scripts that benchmark it against LIME. The main workflow runs model and dataset setups, produces explanations, and evaluates latency, stability, faithfulness, and agreement with SHAP; a separate phase-0 script audits baseline LIME on a movie recommender. The excerpts show the benchmark invoking XAIO and LIME, but omit much of the CLI’s setup and orchestration, so those internal links are intentionally limited to demonstrated relationships.
Main entry points: benchmark_cli.py Benchmark CLI · phase0.py Phase-0 audit · xaio.py XAIO explainer
Components 9
Explainability Benchmarks
- Benchmark CLI
benchmark_cli.py - LIME comparison
benchmark_cli.py - Quality evaluation
benchmark_cli.py - Parameter sweep
benchmark_cli.py
XAIO Runtime
- XAIO explainer
xaio.py
Recommender Baseline
- Phase-0 audit
phase0.py - Movie recommender
phase0.py - Baseline LIME
phase0.py
Other components
- Benchmark operator
Connections 8
- Benchmark operator → Benchmark CLI: runs
benchmark_cli.py - Benchmark CLI → XAIO explainer: invokes
benchmark_cli.py - Benchmark CLI → LIME comparison: compares with
benchmark_cli.py - Benchmark CLI → Quality evaluation: evaluates
benchmark_cli.py - Benchmark CLI → Parameter sweep: sweeps parameters
benchmark_cli.py - Benchmark operator → Phase-0 audit: runs
phase0.py - Phase-0 audit → Movie recommender: loads
phase0.py - Phase-0 audit → Baseline LIME: benchmarks
phase0.py