Convert `np.str_` to regular `str` in Annotations?

[cbrnr]

When creating an Epochs object with "bad" annotations, the non-empty drop log entries currently look like this (tuples of np.str_):

(np.str_('bad_segment'),)

With NumPy 1.x, these entries were regular Python strings, which I think was more readable and more intuitive.

I'm happy to submit a PR to revert the entries to plain strings. Alternatively, we could also consider changing Annotations.description from an np.ndarray with dtype=np.str_ to a plain list[str]. I can't think of a good reason why this field has to be a NumPy array, so I'd really prefer to make this change, which would automatically fix the drop log.

WDYT?

[cbrnr]

OK, vectorized boolean masking only works with NumPy arrays, so we probably might want to keep it. However, the drop_log question still stands.

[drammock]

I believe with numpy 1.x, those were still np.str_ instances, but the repr didn't make that obvious. Regardless, it will be a bit of a pain to switch to using pure list[str] because of e.g., how sorting is implemented

[drammock]

you're better off casting as pure python strings at critical moments (e.g. before save/pickle/etc or before repr)

[cbrnr]

Yes, that's what I suspected. So OK with casting to str for the drop_log for now?

Also, I noticed that the onset, duration, and description attributes are missing in the docs. OK to add in a separate PR?

[drammock]

• cast-to-str for drop log 👍🏻
• add attributes to docs: that's more complicated; in order to show up they'd need to be added as

@property
def onset(self):
   ...

@onset.setter
def onset(self, ...):
   ...

I think it's a worthwhile change but heads up it's maybe a bigger refactor than you were imagining.

[cbrnr]

Alright, I'll create two PRs then! I think the attributes are definitely worth adding, because the whole point of the Annotations class is to provide the three arrays, and these are nowhere to be seen in the docs!

[larsoner]

For the param stuff there is already some related effort underway in #13680

[cbrnr]

Thanks for the heads up @larsoner! I'll see if I can continue from there, or just start from scratch (which might be easier, let's see), or maybe just help get the linked PR finished.