Add constrained_gaussian TOML parser

src/TOMLInterface.jl

@@ -5,7 +5,6 @@ using TOML
 using Distributions
 using EnsembleKalmanProcesses.ParameterDistributions
 
-
 # Exports
 export path_to_ensemble_member
 export get_parameter_distribution
@@ -56,6 +55,10 @@ function get_parameter_distribution(param_dict::Dict, name::AbstractString)
     # Constructing a parameter distribution requires a prior distribution,
     # a constraint, and a name.
     prior = construct_prior(param_dict[name])
+    # If constrained_gaussian, then prior is already a ParameterDistribution
+    if prior isa ParameterDistribution
+        return prior
+    end
     constraint = construct_constraint(param_dict[name])
 
     return ParameterDistribution(prior, constraint, name)
@@ -243,6 +246,35 @@ function get_distribution_from_expr(d::Expr)
 
         return dist_type(dist_args)
 
+    elseif dist_type_symb == Symbol("constrained_gaussian")
+
+        function parse_kwargs(args)
+            kwargs = []
+            for arg in args
+                # Only parse repeats kwarg for now
+                arg.args[1] != :repeats &&
+                    throw(ArgumentError("Keyword argument $(arg.args[1]) can not be parsed from TOML."))
+                push!(kwargs, arg.args[1] => parse(Int64, string(arg.args[2])))
+            end
+            return kwargs
+        end
+
+        kwargs = []
+        index = 2
+        # Non-positional kwargs are the second argument
+        if d.args[2] isa Expr && d.args[2].args[1].head == :kw
+            kwargs = parse_kwargs(d.args[2].args)
+            index += 1
+            # Positional kwargs
+        elseif length(d.args) > 6 && d.args[7].head == :kw
+            kwargs = parse_kwargs([d.args[7]])
+        end
+
+        name, dist_mean, dist_std, lb, ub = d.args[index:(index + 4)]
+        name = string(name)
+        lower_bound, upper_bound = parse.(Float64, string.((lb, ub)))
+        return constrained_gaussian(name, dist_mean, dist_std, lower_bound, upper_bound; kwargs...)
+
     else
         throw(ArgumentError("Unknown distribution type from symbol: $(dist_type_symb)"))
     end

test/TOMLInterface/runtests.jl

@@ -51,6 +51,26 @@ const EKP = EnsembleKalmanProcesses
             [no_constraint(), no_constraint(), bounded_below(-5.0)],
             "uq_param_8",
         ),
+        "uq_param_9" =>
+            ParameterDistribution(Parameterized(Normal(4.0, 0.17881264846405112)), [bounded(0, Inf)], "uq_param_9"),
+        "uq_param_10" => ParameterDistribution(
+            VectorOfParameterized([
+                Normal(4.0, 0.17881264846405112),
+                Normal(4.0, 0.17881264846405112),
+                Normal(4.0, 0.17881264846405112),
+            ]),
+            [bounded(0, Inf), bounded(0, Inf), bounded(0, Inf)],
+            "uq_param_10",
+        ),
+        "uq_param_11" => ParameterDistribution(
+            VectorOfParameterized([
+                Normal(4.0, 0.17881264846405112),
+                Normal(4.0, 0.17881264846405112),
+                Normal(4.0, 0.17881264846405112),
+            ]),
+            [bounded(0, Inf), bounded(0, Inf), bounded(0, Inf)],
+            "uq_param_11",
+        ),
     )
 
     # Get all `ParameterDistribution`s. We also add dummy (key, value) pairs
@@ -65,6 +85,7 @@ const EKP = EnsembleKalmanProcesses
 
     @test_throws ArgumentError get_parameter_distribution(bad_param_dict, "uq_param_baddist")
     @test_throws ArgumentError get_regularization(bad_param_dict, "uq_param_badL")
+    @test_throws ArgumentError get_parameter_distribution(bad_param_dict, "uq_param_bad_constrain_gauss")
 
     for param_name in uq_param_names
         param_dict[param_name]["description"] = param_name * descr

test/TOMLInterface/toml/bad_param.toml

@@ -8,3 +8,6 @@ L2 = 3.5
 prior = "NotParameterized(Normal(-100.0, 20.0))"
 constraint = "no_constraint()"
 L1 = 1.5
+
+[uq_param_bad_constrain_gauss]
+prior = "constrained_gaussian(uq_param_11, 55.47802418037957, 10, 0, Inf; repeats = 3, optim_algo = NelderMead())"

test/TOMLInterface/toml/uq_test_parameters.toml

@@ -40,6 +40,19 @@ constraint = "repeat([no_constraint()], 3)"
 prior = "VectorOfParameterized([Gamma(2.0, 3.0), LogNormal(0.1, 0.1), Normal(0.0, 10.0)])"
 constraint = "[no_constraint(), no_constraint(), bounded_below(-5.0)]"
 
+# Test for `constrained_gaussian` constructor
+# Basic
+[uq_param_9]
+prior = "constrained_gaussian(uq_param_9, 55.47802418037957, 10, 0, Inf)"
+
+# Non-positional kwarg
+[uq_param_10]
+prior = "constrained_gaussian(uq_param_10, 55.47802418037957, 10, 0, Inf; repeats = 3)"
+
+# Positional kwarg
+[uq_param_11]
+prior = "constrained_gaussian(uq_param_11, 55.47802418037957, 10, 0, Inf, repeats = 3)"
+
 # The six parameters below are interpreted as "regular" (non-UQ) parameters, as they
 # they either have no key "prior", or a key "prior" that is set to "fixed"
 [mean_sea_level_pressure]