question about training speed based on a big validation set size about learning-data-manipulation HOT 4 CLOSED

additonal question: "magic_model.update_params(deltas)" must update sample by sample? Is there any method to update batch by batch?

Thank you very much

from learning-data-manipulation.

Re 1: I guess there are two possible ways you can try: (1) sample a subset from the validation set every time. (2) update the augmentation model periodically after multiple steps instead of every single step.

Re 2: Sure it can be batch-by-batch. It's just a designing choice related to coding, no conflict with the method of our paper.

from learning-data-manipulation.

Re 1: I guess there are two possible ways you can try: (1) sample a subset from the validation set every time. (2) update the augmentation model periodically after multiple steps instead of every single step.

Re 2: Sure it can be batch-by-batch. It's just a designing choice related to coding, no conflict with the method of our paper.

Thank you very much!!!!

from learning-data-manipulation.

Re 1: I guess there are two possible ways you can try: (1) sample a subset from the validation set every time. (2) update the augmentation model periodically after multiple steps instead of every single step.

Re 2: Sure it can be batch-by-batch. It's just a designing choice related to coding, no conflict with the method of our paper.

Can you give some idea to run it batch-by-batch? (the option bothers me , in "weighting/classifier.py" line 138:deltas = {name: weights[i] * delta.data for name, delta in deltas.items()})

from learning-data-manipulation.