How can we perfrom pre fitted standard scalar inverse transform on y variable in pipeline

Question

I want to create pipeline which will inverse transform the y variable after model prediction

I have 3 pickle files:

• X variable scalar transform pre fitted object :-X sclr
• Y variable scalar transform pre fitted object :-Y sclr
• svr model object:- model

I imported it to python and now I want create pipeline using this file where input data will get scalar transformed by Xsclr parameter----> then model prediction will happen and then output of model will get inverse transformed using Y sclr object parameter

The implementation would look like

from sklearn.pipeline import Pipeline
pipe = Pipeline(steps=[(StandardScaler(), Xsclr),
                       (SVR(), model),
                       (InverseTransformer(), Ysclr)]
)

I actually I wants to convert this pipeline to ONNX format... Afterwords.

Answer 1

That is not possible in with standard scikit-learn. Scikit-learn Pipelines are not designed to transform y/target variable, they are designed to only work on X/features.

Answer 2

Functionality for transforming regression targets exists, in the TransformedTargetRegressor class; see the docs. With prefitted model and scaler, you might be able to set the "fitted" attributes regressor_ and transformer_.