FIX, DOC: Pipeline methods shouldn't accept just any instance of raw #213
Conversation
Following up from lina-usc#212. I realized that if one did do `my_pipeline.find_outlier_chs(my_raw), then under the hood this method would actually create an epochs instance from my_pipeline.raw ... this would lead to unexpected results if my_raw and my_pipeline.raw are not the same object of memory (i.e. they are different raw objects). Since in our codebase we never do pipeline.find_outlier_chs(raw), I dont think we should support this. Instead, we should either always expect an instanc of mne.Epochs OR We should change the method signature to be def find_outlier_chs(epochs=None) , where if it is None, the method creates epochs from pipeline.raw under the hood.
Codecov ReportAll modified and coverable lines are covered by tests ✅
Additional details and impacted files@@ Coverage Diff @@
## main #213 +/- ##
==========================================
+ Coverage 83.27% 83.43% +0.15%
==========================================
Files 26 26
Lines 1698 1696 -2
==========================================
+ Hits 1414 1415 +1
+ Misses 284 281 -3 ☔ View full report in Codecov by Sentry. |
|
I think this method was probably written that way when we were still passing a def find_outlier_chs(self, epochs, picks="eeg"):
[...]
if epochs is None:
epochs = self.get_epochs(rereference=False, picks=picks) |
- Change API form `inst` to `epochs | None` - if `None`, then make epochs from self.raw - No need check and raise type error now. Co-authored-by: Christian O'Reilly <[email protected]>
pylossless/pipeline.py
Outdated
| else: | ||
| raise TypeError( | ||
| if epochs is None: | ||
| epochs = self.get_epochs(rereference=False, picks=picks) | ||
| "inst must be an instance of mne.Epochs," f" but got {type(inst)}." |
There was a problem hiding this comment.
Oups! I think I left a line that was meant to be removed!
| "inst must be an instance of mne.Epochs," f" but got {type(inst)}." |
There was a problem hiding this comment.
Already fixed in 2a747c2 🙂
|
Alright @christian-oreilly thanks for the review, I think your suggestions made things a lot nicer. FYI please do see 56a2242, I added a line to ensure that we call Before this change, if the user passes in their own epochs, then |
Oops I missed this part. Sure I’m fine with this, would just need to change the function signature to epochs=None so that you can do |
Following up from #212. I realized a couple things...
First of all,
isinstance(inst, mne.io.Raw)is still wrong. It should have beenisinstance(inst, mne.io.BaseRaw).More importantly, I don't think that we should actually support passing a
rawobject to this method...Because for example, if one does
my_pipeline.find_outlier_chs(my_raw), then under the hood this method actually creates an epochs instance frommy_pipeline.raw(viapipeline.get_epochs()), and uses that in outlier detection ...This would lead to unexpected results if
my_rawandmy_pipeline.raware not the same object in memory (i.e. they are different raw objects).Since in our codebase we never do
pipeline.find_outlier_chs(raw), I dont think we should support this.Instead, I think that always expecting an instance of mne.Epochs is OK.
Marking this PR as draft because I need to make sure the new docstring renders OK, but Opening the PR now so that others have time to comment.