Here's the boring model.
If you want to view the charts in Weights and Biases with epoch on the X-axis, you get a message that there is "no data availible". Viewing with the step/global step on the X-axis still works. See the two images below:
With epoch on x-axis

With global_step on x-axis

I suspect this is related to the values that PL sends with the logs, for example this is what the CSVLogger produces:

For some strange reason, there is no epoch value for validation data - the place that should definitely have it...
This bug is extremely annoying if you are, for example, experimenting with different batch sizes - then just comparing steps is meaningless...
* CUDA:
- GPU:
- Tesla T4
- available: True
- version: 10.1
* Packages:
- numpy: 1.18.5
- pyTorch_debug: False
- pyTorch_version: 1.6.0+cu101
- pytorch-lightning: 1.0.2
- tqdm: 4.41.1
* System:
- OS: Linux
- architecture:
- 64bit
-
- processor: x86_64
- python: 3.6.9
- version: #1 SMP Thu Jul 23 08:00:38 PDT 2020
P.S.: Default PL in boring model should be changed to 1.0.2.
I think this might fix the issue https://github.com/PyTorchLightning/pytorch-lightning/pull/4130
I tried, and it does!
Thanks for reporting and reproducing. I need this too!
Most helpful comment
I tried, and it does!