LogExpFunctions.jl
LogExpFunctions.jl copied to clipboard
JuliaStats
Julia package for various special functions based on `log` and `exp`.
This takes advantage of `LogExpFunctions`'s accurate implementations of `log1pexp` and `log1mexp`, combined with negation in the log-odds domain to provide more accurate and less expensive implementations of the function compositions....
Hi, I noticed that R and Python have `logmeanexp`. I think this would be a simple, yet useful addition. I would be willing to make a PR. The basic function...
I've made extensive use of your package in a paper and I would like to include a citation. Please add a CITATION.cff or CITATION.bib file.
```julia julia> using ForwardDiff, LogExpFunctions julia> x = ForwardDiff.Dual(-Inf, 0.0) Dual{Nothing}(-Inf,0.0) julia> logaddexp(x, x) Dual{Nothing}(-Inf,NaN) ``` I think this is related to JuliaDiff/ForwardDiff.jl#481. It is the `a == b` in...
sumlog
This PR adds `sumlog`, a more efficient way to compute `sum(log, x)`. There's more discussion on this on Discourse here: https://discourse.julialang.org/t/sum-of-logs/80370 EDIT: I think we have a good enough understanding...
How about having a version that takes a function to avoid allocation when doing `logsumexp(f.(x))` ? Would you be open to a PR?
Can we have a `logsoftmax` function in this repo? Essentially the same as: https://github.com/FluxML/NNlib.jl/blob/ff3ac6eb807e9b41f46f28f8b3287d19f4b722c7/src/softmax.jl#L107 But I think it would make sense to have it here?
We need to check whether there are numerical accuracy issues or problematic inputs for `Float32` and `Float16`, and if there are, write optimizations for these types.
The README should make it clear that 1. all functions have working forward- and reverse rules, implemented via ChainRulesCore.jl 2. pull requests should implement and test these, too
Metadata
Owner
Metadata
Julia package for various special functions based on `log` and `exp`.