Closed mathematicalmichael closed 1 year ago
Hi @mathematicalmichael,
Thanks for this PR, we'll take a look soon 🙂
Hi @mathematicalmichael ! Thanks for the PR.
Any reason why we would not want to check this at the begging of the __init__
? Also, we could just raise a value error with a simpler message: dataset_name cannot be an empty string
.
Hi @mathematicalmichael ! Thanks for the PR.
Any reason why we would not want to check this at the begging of the
__init__
? Also, we could just raise a value error with a simpler message:dataset_name cannot be an empty string
.
No problem, just wanted to first check this behavior wasnt some intentional feature I'd be breaking by raising an error.
@eugen-ajechiloae-clearml moved the check up. I originally had it where it was because I didn't know if the empty dataset name was a valid thing in the other side of the "if" statement but after closer inspection, it feels relatively reasonable to just not allow people to use dataset names that are empty strings.
Addresses behavior of the project of a dataset being hidden when dataset name is an empty string.
If the dataset project happens to match an existing project, it can be quite jarring to have one's history of experiments suddenly become hidden / make one think they "disappeared" (shown when hidden files enabled in UI, but new tasks can't point to the project name anymore).
If there's an intent behind this behavior (allowing empty dataset names), please let me know and I'd gladly change it to a warning instead.
Perhaps another solution is
dataset_name = None if dataset_name == ""
? or modify the project name under this condition so it's got some ephemeral name to it and prevents the behavior of an existing project getting wiped out.Let me know if a minimal example to reproduce the behavior is desired, I can put one together. Outline of it is:
If the dataset project name differed from the "existing" project, then the original project is untouched, and the dataset-without-a-name is just under a hidden project.
If the dataset name is anything other than an empty string, everything works as anticipated, dataset lives under the right project, nothing is "hidden" that "shouldn't be".
Related Issue \ discussion
https://clearml.slack.com/archives/CTK20V944/p1690230961759919
Patch Description
raises error, prevents user from naming a dataset with an empty string.
Testing Instructions
create a task that creates a dataset. notice the dataset's project is hidden upon completion. if the project names are the same, this makes it look like the task's entire project disappeared from the UI.
Other Information