Slightly refactor MCMCEstimator again.

src/mcmcestimator.jl

@@ -13,9 +13,9 @@ $(TYPEDFIELDS)
 """
 @kwdef struct MCMCEstimator{ST<:Turing.InferenceAlgorithm, SAT<:NamedTuple} <: EstimationMethod
     "Inference algorithm type for MCMC sampling. Defaults to `NUTS`."
-    samplealg::ST = NUTS
-    "kwargs passed to `Turing.sample`. Defaults to `(; )`."
-    sampleargs::SAT = (; )
+    samplealg::ST = NUTS()
+    "kwargs passed to `Turing.sample`. Defaults to `(; drop_warmup=true, progress=false, verbose=false)`."
+    sampleargs::SAT = (; drop_warmup=true, progress=false, verbose=false)
 end
 solvealg(est::MCMCEstimator) = est.samplealg
 solveargs(est::MCMCEstimator) = est.sampleargs
@@ -28,13 +28,12 @@ solveargs(est::MCMCEstimator) = est.sampleargs
     return
 end
 
-function predictsamples(est::MCMCEstimator, f, xs, ysmeas, paramprior::Sampleable, noisemodel::NoiseModel, nsamples;
-                        drop_warmup=true, progress=false, verbose=false,
-                        kwargs...)
+function predictsamples(est::MCMCEstimator, f, xs, ysmeas, paramprior::Sampleable, noisemodel::NoiseModel, nsamples)
+
     chain = with_logger(ConsoleLogger(Warn)) do  # ignore "Info" outputs.
-        alg = solvealg(est)(; solveargs(est)...)
+        alg = solvealg(est)
         sample(bayesianmodel(est, f, xs, maybeflatten(ysmeas), paramprior, noisemodel), alg, nsamples;
-            drop_warmup, progress, verbose, kwargs...)
+               solveargs(est)...)
     end
     d = length(paramprior)
     θsamples = stack([chain[Symbol("θ[$i]")][:] for i in 1:d]; dims=1)