[MRG] Subsampling transformer #259
Conversation
skada/transformers.py
Outdated
| ), | ||
| ) | ||
|
|
||
| idx = self.rng_.choice(X.shape[0], self.n_subsample, replace=False) |
There was a problem hiding this comment.
In practical scenarios, do we need the sampling to be done per-domain? 🤔 It might be the case that the result of sampler omits domains with few(-er) entries.
There was a problem hiding this comment.
In this case do we create a DomainStratifiedSampleTransformer or do we add a parameter? In any Case we should probably use sklearn or skada Splitters here
There was a problem hiding this comment.
I'm always for having separate classes if the use case is clear. Much easier to just open the list of things to spot what you're looking for! Having separate class DomainStratifiedSampleTransformer sounds great.
Codecov ReportAttention: Patch coverage is
Additional details and impacted files@@ Coverage Diff @@
## main #259 +/- ##
=======================================
Coverage 97.15% 97.16%
=======================================
Files 57 59 +2
Lines 6049 6135 +86
=======================================
+ Hits 5877 5961 +84
- Misses 172 174 +2 |
rflamary
changed the title
The main idea is to provide a Subsampler that can be done in a pipeline in order to speedup some methods that do not scale well with number of samples.