Thomas Schill
Thomas Schill
Will need to add citations and update references for alibi-detect's paper and/or documentation before merging the experimental branch.
- [ ] Update README to note the `wilds` setup option - [ ] Update RTD/TOC to include example via `index.rst`
If we have an internal encoder for each detector (or even just labels), then we might want to set up examples so that they decode, e.g. histograms which swap out...
After more discussion, we'll likely leave proper encoding to the user, with examples. One can imagine cases where the incoming data is *already* encoded properly, as it comes out of...
In fixing #96 and #95 , MD3 was left alone. It's still using `DriftDetector` instead of `StreamingDetector`, which has input validation. Supposing we did switch it to `StreamingDetector`, it still...
This doesn't seem to make a consistently big difference, just timing a couple runs of `KDQTreePartitioner.build`. Leaving it for now.
Default values: - [x] CUSUM - [x] PH - [x] ADWIN - [x] DDM - [x] EDDM - [x] LFR - [x] STEPD - [x] CDBD - [x] HDDDM -...
Here's a draft of such a diagram, that excludes concerns like "how fast does it run" and "do you have >=10 features" that would make it an easier decision between...
An alternative would be to switch to a native Python data structure that can implement a linkedlist, instead of the current "by-hand" implementation, which would avoid the unit tests altogether....