Ability to export experiment results to CSV
- Dominant language
- Python
- Stars
- 1.7k
- Forks
- 542
- PR merge metrics
- No merged PRs in 30d
Description
/kind feature
**Describe the solution you'd like**
Currently, there is no way to natively export the results from a Katib experiments.
Given that Katib experiments spawn pods and use up resources (e.g. TFJob experiments with many workers), users might want to delete the experiments to get rid of the pods yet still keep the results. An easy way to do this would be to allow users to download experiment results as a CSV.
A current workaround is to use something like a [Table Capture](https://chrome.google.com/webstore/detail/table-capture/iebpjdmgckacbodjpijphcplhebcmeop?hl=en) extension.
**Anything else you would like to add:**
Part of why this is such an issue for me is that I run a lot of TFJob experiments. This results in some trials failing and the tfjob pods getting stuck at running/using resources. To get rid of these pods, I usually delete the underlying experiments.
It would be nice if there was a way for the pods that run the actual trials could be deleted separately from the experiment so I can keep the results in the UI. I think this might be possible for TFJobs by using `kubectl delete tfjob` but I haven't tried this and it would be cumbersome when I have a hundred trials running. This is kind of out of the scope for this request but just wanted to mention it :)
Contributor guide
Research direction
Start with the Katib experiment-results UI and how experiment results are currently exposed. Consider the TFJob experiment case described in the issue, including what happens to results when experiments or trial pods are deleted. Done means users can download experiment results as CSV while retaining the results they need.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- kubernetes, tensorflow
- Domain
- frontend, machine-learning
- Issue type
- Feature
- Difficulty
- 4/5
- Estimated time
- 3-5 days
- Activity status
- Stale
- Clarity
- Mostly clear
- Newbie friendliness
- 30/100