seldonian.models.sklearn_model.sklearn_predict_vjp

sklearn_predict_vjp(ans, theta, X, model)

Do a backward pass through the Sklearn model, obtaining the Jacobian d pred / dtheta. Must convert back to numpy array before returning.

Parameters:

• ans (numpy ndarray) – The result from the forward pass

• theta (numpy ndarray) – model weights

• X (numpy ndarray) – model features

• model – An instance of a class inheriting from SupervisedSkLearnBaseModel

Return fn:

A function representing the vector Jacobian operator