inverse transform for the predicted/reconstructed values

[dyuliu]

So far, the pipeline will apply a series preprocessing strategies on the raw data to make it ready for ML model.

However, there is a need to observe the predicted/reconstructed values in the original feature space instead of transformed feature space.

Suggestion:

• ORION class records the fitted transformers in the order [T1, T2, T3]
• Add a method to the ORION class called inverse_transform which will apply [i_T3, i_T2, i_T1] to the predicted/reconstructed values

[dyuliu]

Message from @csala:

So far MLPrimitives support functions to reverse the transform: https://github.com/HDI-Project/MLPrimitives/blob/c5208e7bc0f6b8d6afe3aab7bdc0b214bae56f11/mlprimitives/custom/preprocessing.py#L33

Its example usage: https://github.com/HDI-Project/MLPrimitives/blob/reorganize-pipelines/mlprimitives/pipelines/timeseries/forecasting/impute.scale.lstm.json

The only thing here is that this one is a very specific transformation, for min/max scaling, but for other transformations we will not have this architecture ready yet.