David Widmann

The advantage of `true` is that it is the least invasive identity. I assume this was also the motivation behind the definition in base.

Bump :slightly_smiling_face: In particular `invsqrtπ` and `invsqrtquartπ` would be very helpful for the ChainRules definition since it would allow us eg to implement the derivative of `erfcx(x)` as `2 *...

Bump :slightly_smiling_face:

> Since MCMCDiagnosticTools is a somewhat heavy dependency, it's a shame it makes this package less lightweight for non-MCMC applications. Even without the MCMCDiagnosticTools dependency this package is far from...

> Since MCMCDiagnosticTools is a somewhat heavy dependency, it's a shame it makes this package less lightweight for non-MCMC applications. Actually, I think it is not a heavy dependency: https://github.com/TuringLang/ParetoSmooth.jl/issues/44#issuecomment-924419064

> MCMCDiagnosticTools, as we discussed earlier, can be cut out pretty easily. Another way we could cut it out besides what we discussed is just copy/pasting the FFTESS code from...

Sure, the FFTW suggestion was just an additional comment. I haven't benchmarked the precompilation and loading times of MCMCDiagnosticTools but it mainly depends on lightweight packages: https://github.com/devmotion/MCMCDiagnosticTools.jl/blob/main/Project.toml Most of these...

> Another way to cut load times might be switching from FFTESS to another ESS method so we can drop FFTW, but I have no idea how they compare performance/accuracy-wise....

Some benchmarks: https://github.com/TuringLang/MCMCChains.jl/pull/156 and https://github.com/TuringLang/MCMCChains.jl/pull/215 FFT (using FFTW) was by far the slowest method.

This is also why it's neither the default nor recommended :smile: