Nov 08, 2024 aeon Module Meeting (Anomaly Detection)

Present: MM TB SS CH AF PM

[MM] subpackage names
- i.e.
  - "collection" and "series"
  - "whole_series", "point", "subsequence", "correlation"
  - "unsupervised", "supervised", "semi-supervised"
- All of the above could be used probably, but want to avoid something like from anomaly_detection.series.subsequence.unsupervised.deep_learning import
- I'm pretty sure we agreed somthing like this last time, but just confirming:
  - "whole_series"/"collection", "forecasting", "reconstruction", "encoding", "distance", "distribution", "isolation_tree"(?), "deep_learning"
[SS] Deep learning anomaly detection base
- PR #2125
- If self-supervised we have a specific module for that coming up
- Forecasting based, Ali to leave a comment on the PR about structure
[MM] Collection anomaly detection/outlier detection
- OC-SVM first (point-based detection, series estimator)
- there is another student working on ROCK-AD (whole-sequence AD, collection estimator)
- Set up meeting with Andreas regarding whole-series
- MM to make PR on whole-series base class

We then proceed the go off-topic and discuss Docker and numba compilation :)

N/A

N/A

[MM] create issue for streaming (no current plans to implement, from previous meeting)
[MM] create issue for collection AD
[MM] Update AEP with current state and experimental tag discussion
[MM] create issue for subpackages in anomaly detection
[MM] Set up draft PR for whole-series anomaly detection
[TB] Contract Andreas about meeting
[SS] Create issue for Docker image (Numba caching): #2324

[SS] look through survey papers and look if our issues cover all families (good algorithms)
[SS] create issues for PyODAdapter-instances
[MM] create issue for streaming (no current plans to implement)
[MM] create issue for collection AD
[MM] Update AEP with current state and experimental tag discussion

?