Heberto Mayorquin

Which recording are you using? A solution that work for some of our users is to transform to our binary extractor first which should work better with multiprocessing. As @alejoe91...

From your recorder you just do ```python cached_recording = your_recording.save() ``` This will create a copy of your recorder saved in your computer. So there is some extra space that...

I also have the suspicion that some of these problems come from decision of handling the creation and deletion of temporary files.

> I guess there is still the issue or a related one, trying to run built-in spike sorters give an error related to shared memory: ` nbytes = int(np.prod(shape) *...

If I have a vote, I vote to not suppor this. In brief, I think the maintenace burden this will add is not worth the convenience gained. More in detail:...

For discussion but probably does not matter I still feel that some asserts that are about input checking should probably remain asserts. That way, the optimized run of python avoids...

To add to this point, dropping assertions is the only thing that the -O flag does.

Your reasoning makes sense to me. One incentive for people to use asserts instead of errors is that it is easier to comply with code coverage metrics because you don't...

I agree with you.[ It is a bad incentive](https://en.wikipedia.org/wiki/Goodhart%27s_law). I was mentioning this as another reason why the (bad) practice is prevalent besides the synatx being more convenient and that...

> I don't think it does, because in this case you're exposing an object which is already referenced (self._parent_recording). Instead, when registering a recording to a sorting, you're "attaching" a...