Add PseudoObservationsFITC

[patel-zeel]

Added main class and tests for FITC sparse regression.

[wesselb]

Hey @patel-zeel, thanks a lot for putting this together!! Sorry for not getting back earlier to this—I was preoccupied with the AISTATS deadline. Just running the tests now. I'll get back to this with more detailed feedback in the next few days, but on a skim this looks good and basically ready to go. :)

[coveralls]

Pull Request Test Coverage Report for Build 1448930926

• 35 of 35 (100.0%) changed or added relevant lines in 2 files are covered.
• No unchanged relevant lines lost coverage.
• Overall coverage remained the same at 100.0%

---

Totals
Change from base Build 1407415532:	0.0%
Covered Lines:	877
Relevant Lines:	877

---

💛 - Coveralls

[patel-zeel]

No problem at all @wesselb. Thank you for getting back and agreeing for the feedback :)

[wesselb]

Also, if you like, feel free to add an entry to the README which documents the FITC approximation. (I'm also happy to do that myself.)

[patel-zeel]

Hi @wesselb,

Sorry for returning to this after a long time.

I have made the following changes:

• Removed SparseObs and backward compatibility code for it.
• Using PseudoObservations as a parent class for PseudoObsVFE and PseudoObsFITC.
• Using a single @pytest.mark.parametrize for PseudoObsVFE and PseudoObsFITC.
• Corrected example*.py and README to incorporate change from PseudoObsVFE to PseudoObs

Please let me know wherever minor/major changes are required.

[wesselb]

Hey @patel-zeel, absolutely no rush! This is looking great now. :)

Only one comment really: it might not be a good idea to rename PseudoObs to PseudoObsVFE, since that would break existing code depending on the name PseudoObs. I therefore suggest the following renaming:

• PseudoObservations -> AbstractPseudoObservations (to avoid a clash with naming PseudoObservationsVFE back to PseudoObservations)
• PseudoObservationsVFE -> PseudoObservations

With those changed, this seems ready to go! I really appreciate the contribution. :)

[patel-zeel]

Hey @wesselb,

Your suggestion makes perfect sense. I have applied it to the relevant places.

Thank you for your iterative suggestions on this PR. It has been a really good learning experience for me :)

[wesselb]

Thank you @patel-zeel! And no problem at all. :)

This is really great! Your contribution is really appreciated. Super happy to have another approximation implemented.

I've merged the PR. I will release another version of the package in the next few days which includes the FITC approximation. :)

[patel-zeel]

Sounds great! Thank you @wesselb :)