Is there a way to examine the transformed data in an Estimator?

awslabs / gluonts

Probabilistic time series modeling in Python

https://ts.gluon.ai

Apache License 2.0

4.55k stars 748 forks source link

Is there a way to examine the transformed data in an Estimator? #673

Open SkanderHn opened 4 years ago

SkanderHn commented 4 years ago

At some point the estimators are taking sequential time series data and turning it into a structured supervised learning problem. Form the docs and the code, it seems that that is happening during the transformation phase and using the instancesplitter.

Is there a way of pulling that transformed data from the estimator (in order to debug a model that is giving unexpected predictions) as opposed to having to reconstruct the transformation pipeline from scratch?

mbohlkeschneider commented 4 years ago

Hi @SkanderHn,

no, there is no dedicated option to do this.

I recommend to set a breakpoint in your favorite IDE here and use a debugger to inspect the transformation output.