Add init_params keyword argument (#26)

* Update abstractmcmc.jl

* Update interface.jl

* Update src/interface.jl

Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>

* Modified proposal and added basic tests

* Update src/abstractmcmc.jl

Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>

* Correct typos

* Mentionned keyword argument in docs, patch bump and test fix

* Update test/simple.jl

Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>

* Fix tests

* Apply suggestions from code review

Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>

Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
Co-authored-by: David Widmann <devmotion@users.noreply.github.com>

Project.toml

@@ -1,7 +1,7 @@
 name = "EllipticalSliceSampling"
 uuid = "cad2338a-1db2-11e9-3401-43bc07c9ede2"
 authors = ["David Widmann <david.widmann@it.uu.se>"]
-version = "0.4.5"
+version = "0.4.6"
 
 [deps]
 AbstractMCMC = "80f14c24-f653-4e6a-9b94-39d6b0f70001"

docs/src/index.md

@@ -55,6 +55,8 @@ AbstractMCMC.steps(
 gives you access to an iterator from which you can generate an unlimited
 number of samples.
 
+You can define the starting point of your chain using the `init_params` keyword argument.
+
 For more details regarding `sample` and `steps` please check the documentation of
 [AbstractMCMC.jl](https://github.com/TuringLang/AbstractMCMC.jl).
 

src/abstractmcmc.jl

@@ -19,10 +19,14 @@ end
 
 # first step of the elliptical slice sampler
 function AbstractMCMC.step(
-    rng::Random.AbstractRNG, model::AbstractMCMC.AbstractModel, ::ESS; kwargs...
+    rng::Random.AbstractRNG,
+    model::AbstractMCMC.AbstractModel,
+    ::ESS;
+    init_params=nothing,
+    kwargs...,
 )
     # initial sample from the Gaussian prior
-    f = initial_sample(rng, model)
+    f = init_params === nothing ? initial_sample(rng, model) : init_params
 
     # compute log-likelihood of the initial sample
     loglikelihood = Distributions.loglikelihood(model, f)

test/simple.jl

@@ -34,6 +34,11 @@
             @test mean(mean, samples) ≈ μ atol = 0.05
             @test mean(var, samples) ≈ σ² atol = 0.05
         end
+
+        # initial parameter
+        init_x = randn()
+        samples = sample(ESSModel(prior, ℓ), ESS(), 10; progress=false, init_params=init_x)
+        @test first(samples) == init_x
     end
 
     @testset "Scalar model with nonzero mean" begin
@@ -62,6 +67,11 @@
             @test mean(mean, samples) ≈ μ atol = 0.05
             @test mean(var, samples) ≈ σ² atol = 0.05
         end
+
+        # initial parameter
+        init_x = randn()
+        samples = sample(ESSModel(prior, ℓ), ESS(), 10; progress=false, init_params=init_x)
+        @test first(samples) == init_x
     end
 
     @testset "Scalar model (vectorized)" begin
@@ -91,6 +101,13 @@
             @test mean(mean, samples) ≈ μ atol = 0.05
             @test mean(var, samples) ≈ σ² atol = 0.05
         end
+
+        # initial parameter
+        init_x = randn(1)
+        samples = sample(
+            ESSModel(prior, ℓvec), ESS(), 10; progress=false, init_params=init_x
+        )
+        @test first(samples) == init_x
     end
 
     @testset "Scalar model with nonzero mean (vectorized)" begin
@@ -120,5 +137,12 @@
             @test mean(mean, samples) ≈ μ atol = 0.05
             @test mean(var, samples) ≈ σ² atol = 0.05
         end
+
+        # initial parameter
+        init_x = randn(1)
+        samples = sample(
+            ESSModel(prior, ℓvec), ESS(), 10; progress=false, init_params=init_x
+        )
+        @test first(samples) == init_x
     end
 end