add BaytesOptim

Project.toml

@@ -1,7 +1,7 @@
 name = "Baytes"
 uuid = "72ddfcfc-6e9d-43df-829b-7aed7c549d4f"
 authors = ["Patrick Aschermayr <p.aschermayr@gmail.com>"]
-version = "0.3.10"
+version = "0.3.11"
 
 [deps]
 ArgCheck = "dce04be8-c92d-5529-be00-80e4d2c0e197"

src/Baytes.jl

@@ -67,7 +67,7 @@ using ModelWrappers:
     UnflattenStrict,
     UnflattenFlexible
 
-using BaytesMCMC, BaytesFilters, BaytesPMCMC, BaytesSMC
+using BaytesMCMC, BaytesFilters, BaytesPMCMC, BaytesSMC, BaytesOptim
 
 #Utility tools
 import Base: Base, summary
@@ -195,6 +195,11 @@ export
     ParticleMetropolis,
     ParticleGibbs,
 
+    ## Optimizer
+    Optimizer, 
+    OptimConstructor,
+    OptimDefault, 
+
     ## BaytesSMC
     SMC,
     SMCDefault,

src/sampling/utility.jl

@@ -72,13 +72,13 @@ Check if we can capture results from last proposal step. Typically only possible
 """
 function update(datatune::DataTune, temperingadaption::B, smc::SMCConstructor) where {B<:BaytesCore.UpdateBool}
     #!NOTE: If a single SMC constructor is used, we can capture previous results and do not need to update sampler before new iteration.
-    #!NOTE: SMC only updates parameter if UpdateTrue(). Log-target/gradients will always be updated if jitter step applied.
+    #!NOTE: SMC only updates parameter if UpdateTrue(). Log-target/gradients will always be updated if jitter step applied in either case
     return BaytesCore.UpdateFalse()
 end
-function update(datatune::DataTune{<:B}, temperingadaption::BaytesCore.UpdateFalse, mcmc::MCMCConstructor) where {B<:Batch}
-    #!NOTE: If a MCMC constructor is used and no tempering is applied, we can capture previous results and do not need to update sampler before new iteration.
-    #!NOTE: This only holds for Batch data in MCMC case
-    #!NOTE: MCMC updates log target evaluation and eventual gradients if UpdateTrue()
+function update(datatune::DataTune{<:B}, temperingadaption::BaytesCore.UpdateFalse, algorithm::C) where {B<:Batch, C<:Union{MCMCConstructor, OptimConstructor}}
+    #!NOTE: If a MCMC/Optimizer constructor is used and no tempering is applied, we can capture previous results and do not need to update sampler before new iteration.
+    #!NOTE: This only holds for Batch data in MCMC/Optimizer case
+    #!NOTE: MCMC/Optimizer updates log target evaluation and gradients if UpdateTrue()
     return BaytesCore.UpdateFalse()
 end
 function update(datatune::DataTune, temperingadaption, args...)

test/test-construction.jl

@@ -16,6 +16,7 @@ tempermethods = [
 iter = 2
 tempermethod = tempermethods[iter]
 =#
+
 ############################################################################################
 @testset "Sampling, type conversion" begin
     for tempermethod in tempermethods
@@ -323,7 +324,8 @@ using Optim, NLSolversBase
                 )
                 Optimizer(OptimLBFG, :μ)
                 trace, algorithms = sample(_rng, _obj.model, _obj.data, _oc; default = deepcopy(sampledefault))
-
+                ## If single optimizer kernel assigned, can capture previous results
+                @test isa(trace.summary.info.captured, typeof(temperupdate))
 ## Inference Section
                 transform = Baytes.TraceTransform(trace, _obj.model)
                 postmean = trace_to_posteriormean(trace, transform)
@@ -344,17 +346,19 @@ using Optim, NLSolversBase
                 #Check printing commands
                 printchainsummary(trace, transform, Val(:text))
                 printchainsummary(_obj.model, trace, transform, Val(:text))
-#=
-                ## SMC
+
+## SMC via IBIS
                 ibis = SMCConstructor(_oc, SMCDefault(jitterthreshold=0.99, resamplingthreshold=1.0))
                 trace, algorithms = sample(_rng, _obj.model, _obj.data, ibis; default = deepcopy(sampledefault))
-                ## If single mcmc kernel assigned, can capture previous results
+                ## Always update Gradient Result if new data is added
+                    #!NOTE: But after first iteration, can capture results
                 @test isa(trace.summary.info.captured, UpdateFalse)
                 ## Continue sampling
                 newdat = randn(_rng, length(_obj.data)+100)
                 trace2, algorithms2 = sample!(100, _rng, _obj.model, newdat, trace, algorithms)
+                    #!NOTE: But after first iteration, can capture results
                 @test isa(trace2.summary.info.captured, UpdateFalse)
-=#                
+
 ## Combinations
                 trace, algorithms = sample(_rng, _obj.model, _obj.data, mcmc, _oc; default = deepcopy(sampledefault))
                 transform = Baytes.TraceTransform(trace, _obj.model)
@@ -367,25 +371,6 @@ using Optim, NLSolversBase
     end
 end
 
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
 ############################################################################################
 #Utility
 @testset "Utility, maxiterations" begin