I added the logic of active sample for the branch where gp is None and then the samples come from the cache or random box.
This is equivalent to initdesign_vbmc.m in MATLAB.
The difference to MATLAB is that we choose just choose the first N points from the cache when there are more provided points than points to return. In MATLAB KMeans was used for it, but @lacerbi said that it might be changed relatively soon.
I added the documentation to Sphinx and tested all branches of the function. The coverage is 100%.
I added the logic of active sample for the branch where
gp is None
and then the samples come from the cache or random box. This is equivalent to initdesign_vbmc.m in MATLAB.The difference to MATLAB is that we choose just choose the first N points from the cache when there are more provided points than points to return. In MATLAB KMeans was used for it, but @lacerbi said that it might be changed relatively soon.
I added the documentation to Sphinx and tested all branches of the function. The coverage is 100%.