Fiddler’s explainability features require a model on the backend that can generate explanations for you.
A surrogate model is an approximation of your model using gradient boosted trees (LightGBM), trained with a general, predefined set of hyperparameters. It serves as a way for Fiddler to generate approximate explanations without you having to upload your actual model artifact.
A surrogate model will be built automatically for you when you call
You just need to provide a few pieces of information about how your model operates.
- Your model’s task (regression, binary classification, etc.)
- Your model’s target column (ground truth labels)
- Your model’s output column (model predictions)
- Your model’s feature columns
Updated 5 months ago