start: Add new experiments section #4393
Conversation
Link Check ReportThere were no links to check! |
daavoo
force-pushed
the
experiments-trail
branch
from
c6b5bab to
37f5469
Compare
daavoo
force-pushed
the
experiments-trail
branch
from
37f5469 to
98619d2
Compare
This comment was marked as outdated.
This comment was marked as outdated.
This comment was marked as outdated.
This comment was marked as outdated.
This comment was marked as outdated.
This comment was marked as outdated.
|
|
||
| trainer.add_callback(DVCLiveCallback(save_dvc_exp=True)) | ||
| trainer.train() |
There was a problem hiding this comment.
Not a blocker but a follow-up to #4367: should we use the context manager everywhere to be safe?
There was a problem hiding this comment.
should we use the context manager everywhere to be safe?
Not sure.
Given that we are showing the basic callback usage, I am not sure it's worth it to use the context manager here.
Just including it without practical use would make the basic usage look more complicated than it is.
We could show logging additional stuff after the callback but that would complicate the overall example
Thank you! Feel free to open a PR for the other sections and I will start reviewing. |
| If you have [data tracked by DVC](/doc/start/data-management/data-versioning), | ||
| it will be automatically included in the experiment. | ||
|
|
||
| Once you have added [DVCLive] to your code, you can track the changes with git: |
There was a problem hiding this comment.
Not sure I get what this sentence is trying to say. Are we trying to make sure the code gets included with the experiments? If so, maybe it needs to be more explicit, like your code will be included with your experiment if the code file is tracked by git or something like that.
There was a problem hiding this comment.
I am removing it for now and will revisit it if/when we try to make the example more reproducible or tied to the example repo.
|
|
||
| </admon> | ||
|
|
||
| ## Persisting |
There was a problem hiding this comment.
Can we go with apply here? We have tried to put branch first but apply continues to have higher usage. It also doesn't have the awkwardness of a new branch name or leaving a dirty git workspace to clean up, and it's easier to see that everything is in your workspace. Also, I expect branch will seem kind of redundant once you can branch/PR experiments from Studio.
There was a problem hiding this comment.
Can we go with apply here? We have tried to put branch first but apply continues to have higher usage.
I feel like we have gone in circles with this discussion 😅 .
My reasoning is that exp branch is easier to explain/understand (we are in the get started) and less likely to fail or cause unexpected results.
It also doesn't have the awkwardness of a new branch name
or leaving a dirty git workspace to clean up, and it's easier to see that everything is in your workspace.
I don't understand how that is not also true for exp apply.
Also, I expect branch will seem kind of redundant once you can branch/PR experiments from Studio.
I don't understand the redundancy point.
It's like the tabs we already have where you do the same in different contexts.
When it's implemented in Studio, we can just show it in a new tab.
|
Thanks @daavoo! A few questions left, and the links still need to be updated, but I'm ready to approve once all that is done. |
Co-authored-by: Dave Berenbaum <dave@iterative.ai>
Co-authored-by: Restyled.io <commits@restyled.io>
No description provided.