explainable-transformer 0.2.0

Software for debugging transformers

Explainable Transformers

Install by running following command:

pip install explainable-transformers==0.2.0

Usage

from explainable_transformers.text_classification import TextClassificationExplainer

model_path = "cardiffnlp/twitter-roberta-base-sentiment-latest"
explainer = TextClassificationExplainer(model_path=model_path)

inp = ["I like you. I love you."]
shap_values = explainer.get_shap_values(inp)

print(shap_values.base_values)
print(shap_values.values)

Future Scope:

• Want to give an option to enter either a single example or an entire dataset!

  For single example:

  • Output is for local
  • Text Plot is enough

  Entire Dataset: _ Reason: To explain model outputs on a global scale _ Output:

  1. A plot for global values (Required): Waterfall Plot!
  2. Download option for shap values of dataset

• Support for multiple model types and NLP tasks!