search

google-research / uda

Unsupervised Data Augmentation (UDA)
https://arxiv.org/abs/1904.12848
Apache License 2.0
2.17k stars 312 forks source link

Why is this value of "training/unsup/loss" increasing? #73

Open fengyunsanchijian opened 4 years ago

fengyunsanchijian commented 4 years ago

Hi. Why is this value of "training/unsup/loss" increasing? I think this value should be lowered during training. The unsup loss curve

Parameters: INFO:tensorflow:sup batch size: 8 INFO:tensorflow:sup batch size: 8 INFO:tensorflow:unsup batch size: 8 INFO:tensorflow:total sample in a batch: 24

Other parameters are the default. Thanks

airship-explorer commented 4 years ago

Hi.

Why is this value of "training/unsup/loss" increasing? I think this value should be lowered during training. The unsup loss curve

Parameters:

INFO:tensorflow:sup batch size: 8 INFO:tensorflow:sup batch size: 8 INFO:tensorflow:unsup batch size: 8 INFO:tensorflow:total sample in a batch: 24 Other parameters are the default. Thanks

I meet the same situation.