I see that trains is automatically plotting scalars like epoch_acc, epoch_loss. It is not clear where it is picking these from. In one of the runs, I see a epoch_lr but I don't see it in another. Hence, I'm having to manually call report_scalar on the variables of interest. This causes duplication of plots in some case.

figured it out..I can use the logger.report_scalar to do this

So, the usecase is that I am in the middle of developing a model and I have some untracked files, some code which I have not yet pushed in, some code which is not yet upto pep8 standards etc, I dont want it to show up on the dashboard. It will be good to have control over when I want to track git.

Thanks .. I tried both the approaches.

I only see these tensorflow flags and not any of the actual hyperparameters

Thanks..for now.. I just commented some parts of the code to get around this

Thanks Martin...Interesting.. I see the entire agent section missing from the conf file. I guess this happened because I first ran "trains init" and made some manual edits. Then I installed trains-agent on the same machine. "trains-agent init" did not update the already existing file

I think I figured it out. I need to use argparse to set the hyperparameters. Then it logs it automatically