search

worldbank / REaLTabFormer

A suite of auto-regressive and Seq2Seq (sequence-to-sequence) transformer models for tabular and relational synthetic data generation.
https://worldbank.github.io/REaLTabFormer/
MIT License
203 stars 23 forks source link

Is it possible to do iterative training? Load the weight and retrain on new data. #28

Closed vinay-k12 closed 1 year ago

vinay-k12 commented 1 year ago

Instead of loading entire training data, can I train the model on part of it, load the weights and retrain the data?

avsolatorio commented 1 year ago

Hi @vinay-k12, technically, it is possible. However, you should expect discontinuities in the training metrics because the trainer resets the learning rate each time you begin training.

You can always try and see if it works, and if it does, please let me know! 😀

avsolatorio commented 1 year ago

@vinay-k12, I am closing this as inactive. If you've managed to make the iterative training work for you, I'd love to learn more! Thank you. :)