Dev/upgrade prophet 1.1

[jklaise]

Resolves #626.

Went straight to 1.1 as they have nice things:

• No dependency on pystan due to stopped development, the stan backend used now is cmdstanpy
• Wheels provided for all major platforms and Python versions so no installation woes, means we should be able to run Windows tests

I've tested also that the sole example notebooks works as expected.

[codecov-commenter]

Codecov Report

Merging #627 (341af20) into master (6274a96) will not change coverage.
The diff coverage is n/a.

❗ Current head 341af20 differs from pull request most recent head c1531dc. Consider uploading reports for the commit c1531dc to get more accurate results

Additional details and impacted files

@@           Coverage Diff           @@
##           master     #627   +/-   ##
=======================================
  Coverage   77.96%   77.96%           
=======================================
  Files         123      123           
  Lines        8749     8749           
=======================================
  Hits         6821     6821           
  Misses       1928     1928           

Flag	Coverage Δ
macos-latest-3.10.6	74.54% <ø> (ø)
ubuntu-latest-3.10.6	77.34% <ø> (ø)
ubuntu-latest-3.7	77.75% <ø> (ø)
ubuntu-latest-3.8	77.79% <ø> (ø)
ubuntu-latest-3.9	77.79% <ø> (ø)
windows-latest-3.9	74.47% <ø> (ø)

Flags with carried forward coverage won't be shown. Click here to find out more.

[ascillitoe]

• No dependency on pystan due to stopped development, the stan backend used now is cmdstanpy
• Wheels provided for all major platforms and Python versions so no installation woes, means we should be able to run Windows tests

These would both be great!

[ascillitoe]

The test coverage has dropped because the keops tests are not running. I suspect this is because a space is missing in between "Linux" and ] here:

if [ "$RUNNER_OS" == "Linux"]

[jklaise]

Good spot, will change it back. Weitd think is I don't think I edited that line, my editor must've done it automatically which is not great...

[ascillitoe]

Yeh I only know about this issue as ran into the same thing previously. Now I'm wondering if the editor did it last time too...

[jklaise]

I've also updated the changelog (note removal of a couple of extraneous lines from previous releases).

Also note that I've manually checked the save/load functionality still works (we don't have a test for it, happy to add?).

[ascillitoe]

Also note that I've manually checked the save/load functionality still works (we don't have a test for it, happy to add?).

When you say save/load functionality, what are you referring to? Saving and loading OutlierProphet via save_detector and load_detector?

[ascillitoe]

LGTM!

[jklaise]

Also note that I've manually checked the save/load functionality still works (we don't have a test for it, happy to add?).

When you say save/load functionality, what are you referring to? Saving and loading OutlierProphet via save_detector and load_detector?

Yup, that's the one.

[ascillitoe]

When you say save/load functionality, what are you referring to? Saving and loading OutlierProphet via save_detector and load_detector?

Yup, that's the one.

There is actually one test for it in the legacy save/load tests (since outlier detectors are still saved in the legacy format):

alibi-detect/alibi_detect/utils/tests/test_saving_legacy.py

Lines 137 to 141 in 341af20

	if not isinstance(OutlierProphet, MissingDependency):
	detector.append(
	OutlierProphet(threshold=.7,
	growth='logistic')
	)

We'll need to write a new one in saving/tests/test_saving.py once we've updated serialization for outlier detectors.

[arnaudvl]

LGTM!