search

dreamquark-ai / tabnet

PyTorch implementation of TabNet paper : https://arxiv.org/pdf/1908.07442.pdf
https://dreamquark-ai.github.io/tabnet/
MIT License
2.65k stars 488 forks source link

Transfer learning, capability to change structure of model #520

Closed DonCorle0ne closed 11 months ago

DonCorle0ne commented 1 year ago

Feature request

I wonder if it is possible to train a classification model with dataset 1, say 100 features, save the model and use it continue training on dataset 2, but with say 150 features, of which first 100 features are the same as dataset 1.

What is the expected behavior? Be able to save the model and expand the structure to do transfer learning

What is motivation or use case for adding/changing the behavior? This gives user freedom to maintain a base model and explore new intelligence to add in as needed, not forced to train from scratch

How should this be implemented in your opinion?

Are you willing to work on this yourself? yes

Optimox commented 1 year ago

As of my knowledge this is not feasible with any current machine learning algorithm, you need to retrain the model if you change the inputs.