Remove adjoint for fill and fix tests (#203)

* Remove adjoint for `fill` and fix Zygote tests

* Bump version

* Fix some more problems

* Extension of #203: Fix deprecations in test (#204)

* Fix deprecations

* Improve CI (AD): cancel builds and no coverage

* Improve CI (Others): cancel builds and no coverage

* Change parameters to avoid issues with `xlogy`

* Tracker does not like Diagonal(Fill(...))

* Unify CI

* Fix tests

* Update test structure and separate AD better

* Fix tests

* Relax type constraint

* Simplify Zygote tests and use CR

* Improve test design

* Fix typo

* Fix typo

* Replace `unpack` with `_to_vec`

* Fix tests (a bit)

* Fix another test problem

* Fix `_to_vec`

* Fix handling of broken Zygote tests

* Workarounds for `rand_tangent`

* Improvements and fixes for Julia 1.3

* Remove Zygote test hack

.github/workflows/AD.yml → .github/workflows/CI.yml

@@ -1,15 +1,20 @@
-name: AD tests
+name: CI
 
 on:
   push:
     branches:
       - master
   pull_request:
 
+concurrency:
+  # Skip intermediate builds: always.
+  # Cancel intermediate builds: only if it is a pull request build.
+  group: ${{ github.workflow }}-${{ github.ref }}
+  cancel-in-progress: ${{ startsWith(github.ref, 'refs/pull/') }}
+
 jobs:
   test:
     runs-on: ${{ matrix.os }}
-    continue-on-error: ${{ matrix.version == 'nightly' }}
     strategy:
       matrix:
         version:
@@ -19,7 +24,8 @@ jobs:
           - ubuntu-latest
         arch:
           - x64
-        AD:
+        group:
+          - Others
           - ForwardDiff
           - Tracker
           - ReverseDiff
@@ -28,27 +34,42 @@ jobs:
           - version: '1'
             os: macOS-latest
             arch: x64
-            AD: ForwardDiff
+            group: Others
+          - version: '1'
+            os: macOS-latest
+            arch: x64
+            group: ForwardDiff
           - version: '1'
             os: macOS-latest
             arch: x64
-            AD: Tracker
+            group: Tracker
           - version: '1'
             os: macOS-latest
             arch: x64
-            AD: ReverseDiff
+            group: ReverseDiff
           - version: '1'
             os: macOS-latest
             arch: x64
-            AD: Zygote
+            group: Zygote
     steps:
       - uses: actions/checkout@v2
       - uses: julia-actions/setup-julia@v1
         with:
           version: ${{ matrix.version }}
           arch: ${{ matrix.arch }}
+      - uses: actions/cache@v1
+        env:
+          cache-name: cache-artifacts
+        with:
+          path: ~/.julia/artifacts
+          key: ${{ runner.os }}-test-${{ env.cache-name }}-${{ hashFiles('**/Project.toml') }}
+          restore-keys: |
+            ${{ runner.os }}-test-${{ env.cache-name }}-
+            ${{ runner.os }}-test-
+            ${{ runner.os }}-
       - uses: julia-actions/julia-buildpkg@latest
       - uses: julia-actions/julia-runtest@latest
+        with:
+          coverage: false
         env:
-          GROUP: AD
-          AD: ${{ matrix.AD }}
+          GROUP: ${{ matrix.group }}

Project.toml

@@ -1,6 +1,6 @@
 name = "DistributionsAD"
 uuid = "ced4e74d-a319-5a8a-b0ac-84af2272839c"
-version = "0.6.31"
+version = "0.6.32"
 
 [deps]
 Adapt = "79e6a3ab-5dfb-504d-930d-738a2a938a0e"

README.md

@@ -1,7 +1,6 @@
 # DistributionsAD.jl
 
-[![AD tests](https://github.com/TuringLang/DistributionsAD.jl/actions/workflows/AD.yml/badge.svg?branch=master)](https://github.com/TuringLang/DistributionsAD.jl/actions/workflows/AD.yml?query=branch%3Amaster)
-[![Other tests](https://github.com/TuringLang/DistributionsAD.jl/actions/workflows/Others.yml/badge.svg?branch=master)](https://github.com/TuringLang/DistributionsAD.jl/actions/workflows/Others.yml?query=branch%3Amaster)
+[![CI](https://github.com/TuringLang/DistributionsAD.jl/actions/workflows/CI.yml/badge.svg?branch=master)](https://github.com/TuringLang/DistributionsAD.jl/actions/workflows/CI.yml?query=branch%3Amaster)
 
 This package defines the necessary functions to enable automatic differentiation (AD) of the `logpdf` function from [Distributions.jl](https://github.com/JuliaStats/Distributions.jl) using the packages [Tracker.jl](https://github.com/FluxML/Tracker.jl), [Zygote.jl](https://github.com/FluxML/Zygote.jl), [ForwardDiff.jl](https://github.com/JuliaDiff/ForwardDiff.jl) and [ReverseDiff.jl](https://github.com/JuliaDiff/ReverseDiff.jl). The goal of this package is to make the output of `logpdf` differentiable wrt all continuous parameters of a distribution as well as the random variable in the case of continuous distributions.
 

src/flatten.jl

@@ -27,7 +27,6 @@ const flattened_dists = [   Bernoulli,
                             Poisson,
                             Skellam,
                             Arcsine,
-                            Beta,
                             BetaPrime,
                             Biweight,
                             Cauchy,
@@ -40,11 +39,9 @@ const flattened_dists = [   Bernoulli,
                             Exponential,
                             FDist,
                             Frechet,
-                            Gamma,
                             GeneralizedExtremeValue,
                             GeneralizedPareto,
                             Gumbel,
-                            #InverseGamma,
                             InverseGaussian,
                             Kolmogorov,
                             Laplace,
@@ -54,8 +51,6 @@ const flattened_dists = [   Bernoulli,
                             LogitNormal,
                             LogNormal,
                             Normal,
-                            #NormalCanon,
-                            #NormalInverseGaussian,
                             Pareto,
                             PGeneralizedGaussian,
                             Rayleigh,
@@ -64,8 +59,6 @@ const flattened_dists = [   Bernoulli,
                             TriangularDist,
                             Triweight,
                             TuringUniform,
-                            #Truncated,
-                            #VonMises,
                         ]
 for T in flattened_dists
     @eval toflatten(::$T) = true

src/zygote.jl

@@ -1,11 +1,3 @@
-# Zygote fill has issues with non-numbers
-ZygoteRules.@adjoint function fill(x::T, dims...) where {T}
-    return ZygoteRules.pullback(x, dims...) do x, dims...
-        return reshape([x for i in 1:prod(dims)], dims)
-    end
-end
-
-
 ## Uniform ##
 
 ZygoteRules.@adjoint function Distributions.Uniform(args...)