Query Selector Model

[gdevos010]

Fixes #1120.

Summary

This model is a direct adapted from the author's repo. There are also a few more minor documentation fixes mostly recommended by PyCharm.

Other Information

• [x] convert to probabilistic
• [ ] more model documentation
• [ ] tests
• [ ] document license (the apache2.0 license feels too long for the top of the file, so Im not sure the best way).
• [ ] add in layer norm variants.
• [ ] add error handling

[codecov-commenter]

Codecov Report

Merging #1131 (920fee0) into master (df4e2d7) will decrease coverage by 2.73%. The diff coverage is 0.00%.

@@            Coverage Diff             @@
##           master    #1131      +/-   ##
==========================================
- Coverage   93.71%   90.97%   -2.74%     
==========================================
  Files          80       81       +1     
  Lines        8236     8469     +233     
==========================================
- Hits         7718     7705      -13     
- Misses        518      764     +246     

Impacted Files	Coverage Δ
darts/models/forecasting/query_selector.py	0.00% <0.00%> (ø)
darts/timeseries.py	92.23% <0.00%> (-0.07%)	:arrow_down:
...arts/models/forecasting/torch_forecasting_model.py	88.69% <0.00%> (-0.05%)	:arrow_down:
darts/models/forecasting/block_rnn_model.py	98.24% <0.00%> (-0.04%)	:arrow_down:
darts/models/forecasting/nhits.py	98.55% <0.00%> (-0.02%)	:arrow_down:
darts/datasets/__init__.py	100.00% <0.00%> (ø)

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

[gdevos010]

Im currently look for an apartment, I will address these comments when I have a bit more free time

[hrzn]

Im currently look for an apartment, I will address these comments when I have a bit more free time

Sure, no rush

[gdevos010]

@hrzn I originally wanted to add it because it looked to do reasonably well on benchmarks. I have run a bunch of tests and haven't been impressed. I mostly get forecasts that look like these plots below. I have done some tuning and cant seem to get it to "learn" the data. I will move the typos I made to a new branch.

[hrzn]

@hrzn I originally wanted to add it because it looked to do reasonably well on benchmarks. I have run a bunch of tests and haven't been impressed. I mostly get forecasts that look like these plots below. I have done some tuning and cant seem to get it to "learn" the data. I will move the typos I made to a new branch.

OK, thanks for investigating! In general, I think we should indeed favour models that either perform very well, or are accompanied by a solid publication demonstrating the value. Your results are not great, however looking at them I would tend to think they look more like a bug in the implementation than an issue with the model itself. We can still park this PR/branch until someone has more time to investigate.

[hrzn]

@gdevos010 I will close this one for now, but let's be open about re-opening if the time comes later on. Hope that's OK for you.