Support passing prior_predictive without prior to from_mcmcchains (#152)

* Convert prior predictive even if prior not provided

* Add test

* Increment version number

Project.toml

@@ -1,7 +1,7 @@
 name = "ArviZ"
 uuid = "131c737c-5715-5e2e-ad31-c244f01c1dc7"
 authors = ["Seth Axen <[email protected]>"]
-version = "0.5.10"
+version = "0.5.11"
 
 [deps]
 Conda = "8f4d0f93-b110-5947-807f-2305c1781a2d"

src/mcmcchains.jl

@@ -268,6 +268,14 @@ function from_mcmcchains(
             ),
         )
         concat!(all_idata, prior_idata)
+    elseif prior_predictive !== nothing
+        pre_prior_predictive_idata = convert_to_inference_data(
+            prior_predictive; eltypes=eltypes, kwargs...
+        )
+        concat!(
+            all_idata,
+            InferenceData(; prior_predictive=pre_prior_predictive_idata.posterior),
+        )
     end
 
     for (group, group_data) in [

test/test_mcmcchains.jl

@@ -250,6 +250,17 @@ end
         @test "z" ∈ keys(dimdict(idata.predictions_constant_data))
     end
 
+    # https://github.com/arviz-devs/ArviZ.jl/issues/146
+    @testset "prior predictive w/o prior" begin
+        nvars, nchains, ndraws = 2, 4, 20
+        chns = makechains(nvars, ndraws, nchains)
+        prior_predictive = randn(nchains, ndraws, 1)
+        idata = from_mcmcchains(chns; prior_predictive=prior_predictive)
+        test_chains_data(chns, idata, :posterior, names(chns))
+        @test :prior_predictive ∈ ArviZ.groupnames(idata)
+        @test idata.prior_predictive.x.values ≈ prior_predictive
+    end
+
     @testset "missing -> NaN" begin
         rng = MersenneTwister(42)
         nvars, nchains, ndraws = 2, 4, 20