zoj613

Also worth noting that this behaviour happens even if we call `polyagamma_pdf` without the `return_log` flag:  Also note that for larger values of the exponential tilting parameter (e.g `z>=5`),...

@martinjankowiak I submitted a re-write at #81 that addresses part of the issue and ensures nans are not returned. However the instability still persists for large `h` when `z=0`, leading...

I considered a saddle point approximation of the density as a viable solution for large values of `(x, h)`. I came to the conclusion that it might not be the...

any updates on the development of this project?

@brandonwillard where would the implementation be put? (basic.py vs math.py vs nlinalg.py). Also which of the input parameters should be ignored for the aesara implementation?

I can have a go at this. Which version do you prefer? I'm thinking this would be good to add: https://github.com/NTimmons/FastActivations.jl/blob/aaa61f84dfcd75825b20676011355189cb497d8d/src/SigmoidFittedApproximations.jl#L407-L465 Is there a benchmark suite to verify said claims...

Could you link please to the paper? I don't see it mentioned there.

Same here. Did you manage to solve this @4n4h3l ?

> Need to prioritize this given that TDS 5 is now the stable release and Unidata production servers are now using it. > > For instance, today running this: >...

> Is there a scipy bug report or pull request where this idea had been rejected? No there isn't. It is my guess based on scipy policies about new code....