search

KIT-IAI / pyWATTS

pyWATTS: Python Workflow Automation Tool for Time-Series
MIT License
39 stars 9 forks source link

Training Evaluation when using pyTorch #32

Open marianturowski opened 3 years ago

marianturowski commented 3 years ago

What is your feature request related to?

Please describe the module/problem

Based on feedback from @olineumann: Currently, one can only do random guessing on the number of epochs, batch size, ... because there is no feedback on how the training performs. Therefore, as a user of PyWATTS, it would be useful to be able to review the training process of pyTorch models, for example, in terms of a train/test loss curve.

Describe the solution you'd like

Maybe there is a way to add terminal printing? Maybe there is a way to plot train/test loss curves? Maybe there is a way to pass an individual method for evaluation after each epoch?

benHeid commented 3 years ago

Perhaps it can be a combined solution with #34 .

Perhaps, it would be possible to not only return the resulted dataarray but also additionally summary informations that are handled also by the step.