When I launch multiple experiments across a parameter grid using hydra multirun, the scalars that were auto connected from tensorflowx act as if they were continuing from the same experiment in serial rather than different experiments. This makes it difficult to compare scalars across different experiments.

I've tried explicitly resetting the iteration account back to 0 across runs using set_initial_iteration with no luck

So what's weird is that I'm explicitly closing the task but the separate experiments are not starting back at iteration 0

I'm using 0.17.4

Ah, thanks for looking into it!

The newest rc does indeed fix the issue, thank you!

The manual color change seems to be locked to all 3 lines for me
i.e. I can't manually set the lines to be different colors

re:hashing str - Ah gotcha!

Ah, to clarify I'm hoping to have different colors for each line to better differentiate between them. Right now they all seem to be locked to the same color e.g all blue, or all red

Yes, I'm using the community server. So I took a look at one of the examples experiments and I am able to individually change the color of different lines e.g the "scalar reporting" example. I wonder if it has to do with how I called the reporting function for my experiment?