Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

"Run picker" #414

Open
haeusser opened this issue Aug 24, 2017 · 1 comment
Open

"Run picker" #414

haeusser opened this issue Aug 24, 2017 · 1 comment

Comments

@haeusser
Copy link

Hi all,

I've been using Tensor{Flow,Board} for quite a while.

One tool that I've been missing is a "picker" that allows you to select runs from graphs (scalars chart).

User story:
I open TB, unfold plots of my (many) runs and I want to follow up with a particular one.
In the graph, I click on the run that I am interested in.
This copies the run name in the clipboard and I can filter or do whatever else with it.
I would like to get a mapping from chart to run name. In many situations you have a ton of runs that potentially have un-handy names. Such a picker would allow to easily select runs based on their performance.

Version 2:
I can chose shortlists from the plots.

Version 3:
From a chart, I can automatically select the top/bottom N runs. For example, with one click I get the name of the run that logged the highest accuracy or lowest test error.

It would be awesome if we can get this implemented somehow. I'd be happy to help. What are your thoughts?

Cheers,
Philip

@psybuzz
Copy link
Contributor

psybuzz commented Jan 23, 2021

Thank you for the feature request! We appreciate these specific use cases.

I think there are 2 core limitations of the scalars dashboard that make the use cases you mention a challenge.

  • The tooltip appearing over scalar charts on hover has no easy affordance for selecting a run/copying its name/sorting to see top N
  • After identifying "best" runs in a chart, there is no way to "hide everything else" and view only a shortlist in each plot

The first issue is something related to #1019. There isn't yet a solution for that other issue, but I could imagine that if TensorBoard showed a persistent table of values, it would be straightforward to filter/sort/hide columns of interest and copy text.

The second issue is something that the team would like to work on. Being able to group runs might provide a reasonable affordance for creating a subset of the "best runs". There is a lot of UX design work to be done on the topic of run grouping, in my opinion.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

2 participants