Anomaly Detection (anomaly model, scorer, detector, aggregator)

[julien12234]

Fixes #.

Summary

Other Information

[review-notebook-app]

Check out this pull request on 

See visual diffs & provide feedback on Jupyter Notebooks.

---

Powered by ReviewNB

[hrzn]

Just a first round of comments, mostly about formatting for now :)

[codecov-commenter]

Codecov Report

Base: 93.83% // Head: 93.93% // Increases project coverage by +0.10% 🎉

Coverage data is based on head (82b65cb) compared to base (919f214).
Patch coverage: 94.94% of modified lines in pull request are covered.

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #1256      +/-   ##
==========================================
+ Coverage   93.83%   93.93%   +0.10%     
==========================================
  Files          94      122      +28     
  Lines        9673    10709    +1036     
==========================================
+ Hits         9077    10060     +983     
- Misses        596      649      +53     

Impacted Files	Coverage Δ
darts/ad/anomaly_model/anomaly_model.py	80.76% <80.76%> (ø)
darts/ad/utils.py	90.86% <90.86%> (ø)
darts/ad/anomaly_model/filtering_am.py	91.93% <91.93%> (ø)
darts/ad/anomaly_model/forecasting_am.py	93.45% <93.45%> (ø)
darts/ad/scorers/scorers.py	95.13% <95.13%> (ø)
darts/ad/aggregators/aggregators.py	95.77% <95.77%> (ø)
darts/ad/detectors/detectors.py	96.07% <96.07%> (ø)
darts/ad/__init__.py	100.00% <100.00%> (ø)
darts/ad/aggregators/__init__.py	100.00% <100.00%> (ø)
darts/ad/aggregators/and_aggregator.py	100.00% <100.00%> (ø)
... and 23 more

Help us with your feedback. Take ten seconds to tell us how you rate us. Have a feature suggestion? Share it here.

☔ View full report at Codecov.
📢 Do you have feedback about the report comment? Let us know in this issue.

[hrzn]

I've made a round of comments.

Overall, I think it's a great start and taking shape. Here are the most important points:

• Please, think of what happens both for (i) multivariate series and (ii) multiple series. It seems to me that the code is not general enough in few places about that.
• Unless I'm missing something, I think most scorers (e.g. L2, but also KMeans and others) should act in the original D-dimensional space for D-dimensional multivariate series, and always return univariate series (I don't think it's really done like that currently).
• Somehow, I would expect KMeansScorer and other similar scorers to work over certain rolling windows (kind of what you have for Wasserstein). This is probably where the characteristic time length should play (rather than in some functions keeping e.g. the max values over certain windows as a postprocessing step for all scorers - I would suggest removing this).
• I would let the scorers compute the distance themselves if they need it, rather than feeding them with (some variants) of residuals.
• I haven't looked at Detectors yet
• I have pushed a couple of commits with small improvements (mainly to docstrings), but more corrections need to be made there

[eliane-maalouf]

AD is a great addition to the Darts @julien12234 and @hrzn . Thanks for this colossal work :)

[hrzn]

🥳