StochasticAD.jl

StochasticAD.jl copied to clipboard

Zygote.gradient returns nothing when differentiating vectorized sampling.

1

Hi there, first of all, thanks for this great package. I'm trying to understand how StochasticAD integrates with Zygote and I ran into a problem when trying to differentiate a...

arnauqb

One-hot Encoding

2

Hi there! I've been trying to implement a one hot encoding using StochasticAD. So far, I've failed 🥲. I think it essentially boils down to [this TODO](https://github.com/gaurav-arya/StochasticAD.jl/blob/d08156a397f20477f25bf6bef86d7257e7160f08/src/general_rules.jl#L235) in the src....

AlCap23

The internal global RNG is undesirable because it makes pruning non-deterministic even if the user fixes a seed for the RNGs they control. We could fix this by splitting off...

gaurav-arya

Make propagate more like a monadic bind by supporting stochastic triple creating functions

6

x-ref #128. @GuusAvis example: ```julia using StochasticAD using Distributions function f(value_1, value_2, rand_var) if value_1 < value_2 return (value_1 + rand(rand_var), value_2) else return (value_1, value_2 + rand(rand_var)) end end...

gaurav-arya

Added two new tests of propagate that check if finite perturbations returned by functions upon normal calls are treated correctly by `propagate`. 1. Adding two stochastic triples together, like in...

GuusAvis

Added type testing for some kwargs in `propagate`, and adjusted a test that I think was already failing because of a wrong type but now it no longer fails quietly....

GuusAvis

stochastic triples not passed through propagate lose their perturbations

7

Hello! I have been running into the following problem. When I use `propagate` to pass a stochastic triple to a function, but in that function there is a second stochastic...

GuusAvis

This pull request sets the compat entry for the `Enzyme` package to `0.12`. This keeps the compat entries for earlier versions. Note: I have not tested your package with this...

github-actions[bot]

`propagate` to add two stochastic triples together

5

Hello! I'm trying to use the `propagate` function to extract derivatives from some Monte Carlo simulations. However, I noticed that the function does not behave as I would expect in...

GuusAvis

Fixes #125

Krastanov