Dave Berenbaum
Dave Berenbaum
> have to manually report certain metrics after the train loop is done Like best accuracy? It doesn't seem like keras makes it easy to retrieve this info, but maybe...
Can you easily separate the `log_image` changes from the rest? I think that's easy to merge, but the other part seems a little more divisive . > This is a...
> Should `make_summary` be called `write_summary` or just `log_summary`? Hmm, no strong opinion, but should this method also be either made private or documented?
> The idea is that, for no-step workflow, we document the following (instead of calling `live.log_metric`): > > ```python > live = Live() > > live.summary["foo"] = 1 > live.make_summary()...
Discussed with @daavoo and agreed to: 1. Keep using `live.log()` as the primary workflow for step and no-step scenarios but without any special syntax for no-step scenarios. This means no-step...
Sorry, forgot to approve this one.
I would suggest we prioritize at the same time as #305 since it's part of the same story (auto logging everything from the framework callbacks).
Should this also create a `.dvc` file to track the model output? In a DVC pipeline, it would be necessary to either ignore this step or find some alternate integration...
Also, this should be integrated into callbacks so that it's easy to save models automatically. Related: #300.
> a related issue > > * [`dvclive` integration? mlem#2](https://github.com/iterative/mlem/issues/2) Good point, let's discuss there since you both already covered most of what's here and more. Edit: Although so far,...