Table 1.
Summary of available open-source XAI tools
| Toolkit | Data Explanations | Directly Interpretable | Self-explaining | Local Post Hoc Explanation | Global Post Hoc Explanation | Explaina- bility Metrics | URL Links |
|---|---|---|---|---|---|---|---|
| AIX 360 | X | X | X | X | X | X | http://aix360.mybluemix.net |
| Alibi | X | https://github.com/SeldonIO/alibi | |||||
| Skater | X | X | X | https://oracle.github.io/Skater/ | |||
| H2O | X | X | X | https://github.com/h2oai/mli-resources | |||
| InterpretML | X | X | X | https://github.com/interpretml/interpret | |||
| EthicalML-XAI | X | https://github.com/EthicalML/xai | |||||
| DALEX | X | X | https://modeloriented.github.io/DALEX/ | ||||
| tf-explain | X | X | https://github.com/sicara/tf-explain | ||||
| iNNvestigate | X | https://github.com/albermax/innvestigate | |||||
| modelStudio | X | X | X | X | https://bit.ly/3uOnU5y | ||
| ELI5 | X | X | X | https://github.com/TeamHG-Memex/eli5 | |||
| Iml | X | X | X | https://bit.ly/3iBv8Vx |
Coverage is shown along several explainability dimensions: (1) data explanations are provided through data distributions and enable case-based reasoning; (2) directly interpretable refers to a model that inherently provides information about features driving predictions at both global and local levels; (3) in contrast, self-explaining models provide local explanations but may not be globally interpretable; (4) local post hoc explainers can provide explanations around particular data points for black-box models in a post hoc manner; (5) whereas global post hoc explainers provide the same at a global/model level; and (6) explainability metrics cover several state of the art metrics to quantify the explainers/models around several dimensions of explainability.