EdwardRaff
EdwardRaff
I am not going to introduce a breaking change like renaming a file without understanding the situation. I really appreciate your interest, but I am simply not making any changes...
I read the wiki page, and did not find it to be very comprehensible. But what you've just written is much more cogent. However, it also does not match what...
If you are trying to benchmark and compare, I totally understand the frustration. But please also understand from my side, I have no clarity one what is the real PAM...
I have some upcoming work that I _think_ this might solve, but the documentation is very sparse. Are the `subposteriors` the states from previous NUTS runs? Can it work with...
Thanks! You can put these typos you find in one doc it will be easier for me to correct when the published OKs a 2nd edition :)
Adding another request for CPU atomics. I'm not sure how multi-dimensional cases are handled, but abstractly if there was an `atomic_add(arr, indx, val)` `atomic_compare_and_swap(arr, index, expected, new)` and an `atomig_get(arr,...
@arthurp that looks interesting. Could you elaborate further on when it may not work? If I pass a numpy array in, the original code has the pointer/ref count, I'm not...
Bumping that the lack of CPU atomics in Numba is a huge pain point/blocker for implementing any interesting parallel algorithms.
What version of Cython were you using? I want to do some quick checks before accepting it.
I apparently forgot about this. @bobo-le does edmcman's patch fix your issue?