Typo in the formula of Bichon and Ranjan criteria, other points and question

[mdelozzo]

Thanks for this nice package! I've got a few points.

The LaTeX formula for Bichon and Ranjan criteria seems wrong. In Equation (31) of the citer paper, we can read the conditionned GP $\xi(x)$ instead of its mean $m(x)$, as currently written in the docstring.

Other points:

• the rendering of the AEI is broken; the LaTeX line should be surrounded by empty lines in the docstring.
• $f$ is both the function to approximate and the conditionned GP in the documentation; two separate notations would be clearer,
• a section about notations would be useful, and perhaps a page listing the criteria (in addition to the API pages).

Question: is it possible to use specific basis functions for the GP mean?

[uri-granta]

Thank you for spotting and raising these!

And yes, it is possible to use specific basis functions: there are various built-in functions and you can define your own, as described at https://gpflow.github.io/GPflow/develop/notebooks/getting_started/mean_functions.html.

[uri-granta]

@hstojic could you please check the bichon_ranjan_criterion docstring issue mentioned above?

[hstojic]

ok, I'll check it out - while we are at it, @uri-granta probability_below_threshold also has a broken equation, and there is something weird in batch_expected_improvement with _delta method docstring

[mdelozzo]

Thanks @uri-granta for your link. I'll take a closer look at the API documentation.