orion

orion copied to clipboard

**Describe the bug** When restoring the state of the EvolutionES algorithm, when it already has trials, the new trials suggested by the algorithm have the same params, but a different...

lebrice

Suggested improvements to the `Trial` class

1

Here are some of my thoughts about the `Trial` class: - Creating `Trial` objects is too cumbersome atm in my opinion. To create a `Trial`, you invariably have to go...

lebrice

Flaky test. Seems to be failing randomly for some reason, on python 3.8 only. Unable to reproduce the issue locally.

lebrice

Remove most uses of `collections.OrderedDict`

1

Since we dropped support for python 3.6, and python dictionaries are ordered since python 3.7, we can now replace most uses of `collections.OrderedDict` with the built-in `dict`, and also remove...

lebrice

Support maximizing objective in `workon`, `report_objective`, etc.

1

Instead of asking users to do something, we should do it ourselves. This is the kind of super basic thing that should absolutely be supported natively by `Orion`, without having...

lebrice

Convert modules to use typing annotation.

1

We need to first list the modules to track the progress of conversion. We should also look for a way to document arguments without specifying types twice.

bouthilx

See https://github.com/automl/SMAC3/issues/829#issuecomment-1088860861

bouthilx

Warnings about non-tuple indexing or np.object dtypes happen during tests. Would be nice to get rid of those.

lebrice

From #860 > Oh and while we're at it, I suggest we add a self.max_trials: Optional[int] = None attribute in BaseAlgorithm. > Lots of tests forcefully set this attribute on...

bouthilx

See short discussion here: https://github.com/Epistimio/orion/pull/867 (resolved review)

bouthilx