TensorBoard not plotting all available scalar runs

Viewed 32

I'm running some tests with different LSTM models using Keras and TensorFlow. I'm also using TensorBoard to monitor the model's losses.

In one of my tests, I noticed that even though all my runs were being displayed (and were selected) in the "Runs" section, not all of them were plotted. Out of a total of 18 runs, only 7 showed up in the plots (https://imgur.com/5UKLXcF). I unselected these 7 and the plots were then blank — even though 11 runs were still selected (https://imgur.com/6iDnWNE).

I took note of these runs and used TensorBoard again only on 2 of them, which I knew hadn't worked before. This time the plots showed up normally (https://imgur.com/3VamZnF).

This also doesn't seem consistent at all, I've tried opening the full 18 again multiple times and in some of them the plots that showed were different from the previous 7, but never the full 18 at the same time. Anyone know what is going on? Is there a way to fix this?

0 Answers
Related