gh-1131: Attempt to make reg tests stable#1132
Conversation
This reverts commit f54e805.
| gls: AngularPowerSpectra = [xpb.ones(10) for _ in range(nth_triangular_number)] | ||
| nside = 16 | ||
| gls: AngularPowerSpectra = [xpb.zeros(10) for _ in range(nth_triangular_number)] | ||
| nside = 32 |
There was a problem hiding this comment.
Why changing from ones to zeros?
There was a problem hiding this comment.
This prevents the exception "covariance matrix is not positive definite" from being raised, thus allowing us to consume the full generator.
There was a problem hiding this comment.
Why weren't we doing that before?
There was a problem hiding this comment.
I'm not sure, I guess it depends if passing all zeros simulates a real situation worth benchmarking?
There was a problem hiding this comment.
It doesn't sound obvious. How long does it take? Should we consult Nicolas on this?
There was a problem hiding this comment.
Probably. @ntessore. Does making gls all zeros make this benchmark not realistic?
There was a problem hiding this comment.
No, it will produce Gaussian random draws which are all zero, but the amount of work done is still the same.
Another option is
[xpb.ones(10) if i == j else xpb.zeros(10) for i in range(10) for j in range(i, -1, -1)]
There was a problem hiding this comment.
Okay, So then I guess we are okay to merge? (assuming an approving review)
There was a problem hiding this comment.
I think I'd prefer Nicolas' suggestion as it reflects a real world case
There was a problem hiding this comment.
I've updated to use the new example
|
Still got in #1133 |
3 participants
Description
test_generate[_grf]testsCloses: #1131
Checks