Merge pull request #164 from ilabcode/test

removed category means and noisy binary

docs/julia_files/tutorials/classic_binary.jl

@@ -21,8 +21,6 @@ inputs = CSV.read(data_path * "classic_binary_inputs.csv", DataFrame)[!, 1];
 
 # Create an HGF
 hgf_parameters = Dict(
-    ("u", "category_means") => Real[0.0, 1.0],
-    ("u", "input_precision") => Inf,
     ("xprob", "volatility") => -2.5,
     ("xprob", "initial_mean") => 0,
     ("xprob", "initial_precision") => 1,
@@ -54,8 +52,6 @@ plot_trajectory(agent, ("xvol", "posterior"))
 # Set fixed parameters
 fixed_parameters = Dict(
     "action_noise" => 0.2,
-    ("u", "category_means") => Real[0.0, 1.0],
-    ("u", "input_precision") => Inf,
     ("xprob", "initial_mean") => 0,
     ("xprob", "initial_precision") => 1,
     ("xvol", "initial_mean") => 1,

docs/julia_files/user_guide/fitting_hgf_models.jl

@@ -45,8 +45,6 @@ using HierarchicalGaussianFiltering
 # We will define a binary 3-level HGF and its parameters
 
 hgf_parameters = Dict(
-    ("u", "category_means") => Real[0.0, 1.0],
-    ("u", "input_precision") => Inf,
     ("xprob", "volatility") => -2.5,
     ("xprob", "initial_mean") => 0,
     ("xprob", "initial_precision") => 1,
@@ -86,8 +84,6 @@ plot_trajectory!(agent, ("xbin", "prediction"))
 # Set fixed parameters. We choose to fit the evolution rate of the xprob node. 
 fixed_parameters = Dict(
     "action_noise" => 0.2,
-    ("u", "category_means") => Real[0.0, 1.0],
-    ("u", "input_precision") => Inf,
     ("xprob", "initial_mean") => 0,
     ("xprob", "initial_precision") => 1,
     ("xvol", "initial_mean") => 1,

docs/julia_files/user_guide/utility_functions.jl

@@ -62,8 +62,6 @@ agent_parameter = Dict("action_noise" => 0.3)
 #We also specify our HGF and custom parameter settings:
 
 hgf_parameters = Dict(
-    ("u", "category_means") => Real[0.0, 1.0],
-    ("u", "input_precision") => Inf,
     ("xprob", "volatility") => -2.5,
     ("xprob", "initial_mean") => 0,
     ("xprob", "initial_precision") => 1,

src/ActionModels_variations/utils/set_parameters.jl

@@ -40,12 +40,6 @@ function ActionModels.set_parameters!(
         )
     end
 
-    #If the param is a vector of category_means
-    if param_value isa Vector
-        #Convert it to a vector of reals
-        param_value = convert(Vector{Real}, param_value)
-    end
-
     #Set the parameter value
     setfield!(node.parameters, Symbol(param_name), param_value)
 

src/create_hgf/hgf_structs.jl

@@ -340,8 +340,6 @@ end
 Configuration of parameters in binary input node. Default category mean set to [0,1]
 """
 Base.@kwdef mutable struct BinaryInputNodeParameters
-    category_means::Vector{Union{Real}} = [0, 1]
-    input_precision::Real = Inf
     coupling_strengths::Dict{String,Real} = Dict{String,Real}()
 end
 

src/premade_models/premade_hgfs/premade_binary_2level.jl

@@ -11,8 +11,6 @@ function premade_binary_2level(config::Dict; verbose::Bool = true)
 
     #Defaults
     spec_defaults = Dict(
-        ("u", "category_means") => [0, 1],
-        ("u", "input_precision") => Inf,
         ("xprob", "volatility") => -2,
         ("xprob", "drift") => 0,
         ("xprob", "autoconnection_strength") => 1,

src/premade_models/premade_hgfs/premade_binary_3level.jl

@@ -13,8 +13,6 @@ This HGF has five shared parameters:
 "coupling_strengths_xprob_xvol"
 
 # Config defaults:
- - ("u", "category_means"): [0, 1]
- - ("u", "input_precision"): Inf
  - ("xprob", "volatility"): -2
  - ("xvol", "volatility"): -2
  - ("xbin", "xprob", "coupling_strength"): 1
@@ -28,8 +26,6 @@ function premade_binary_3level(config::Dict; verbose::Bool = true)
 
     #Defaults
     defaults = Dict(
-        ("u", "category_means") => [0, 1],
-        ("u", "input_precision") => Inf,
         ("xprob", "volatility") => -2,
         ("xprob", "drift") => 0,
         ("xprob", "autoconnection_strength") => 1,

src/update_hgf/node_updates/binary_state_node.jl

@@ -99,13 +99,8 @@ function calculate_posterior_precision(node::BinaryStateNode)
         child = node.edges.observation_children[1]
 
         #Simple update with inifinte precision
-        if child.parameters.input_precision == Inf
-            posterior_precision = Inf
-            #Update with finite precision
-        else
-            posterior_precision =
-                1 / (node.states.posterior_mean * (1 - node.states.posterior_mean))
-        end
+        posterior_precision = Inf
+
 
         ## If the child is a category child ##
     elseif length(node.edges.category_children) > 0
@@ -141,30 +136,7 @@ function calculate_posterior_mean(node::BinaryStateNode, update_type::HGFUpdateT
             #Set the posterior to missing
             posterior_mean = missing
         else
-            #Update with infinte input precision
-            if child.parameters.input_precision == Inf
-                posterior_mean = child.states.input_value
-
-                #Update with finite input precision
-            else
-                posterior_mean =
-                    node.states.prediction_mean * exp(
-                        -0.5 *
-                        node.states.prediction_precision *
-                        child.parameters.category_means[1]^2,
-                    ) / (
-                        node.states.prediction_mean * exp(
-                            -0.5 *
-                            node.states.prediction_precision *
-                            child.parameters.category_means[1]^2,
-                        ) +
-                        (1 - node.states.prediction_mean) * exp(
-                            -0.5 *
-                            node.states.prediction_precision *
-                            child.parameters.category_means[2]^2,
-                        )
-                    )
-            end
+            posterior_mean = child.states.input_value
         end
 
         ## If the child is a category child ##

src/utils/get_surprise.jl

@@ -103,40 +103,15 @@ function get_surprise(node::BinaryInputNode)
         parents_prediction_mean += parent.states.prediction_mean
     end
 
-    #If the input precision is infinite
-    if node.parameters.input_precision == Inf
-
-        #If a 1 was observed
-        if node.states.input_value == 0
-            #Get surprise
-            surprise = -log(1 - parents_prediction_mean)
-
-            #If a 0 was observed
-        elseif node.states.input_value == 1
-            #Get surprise
-            surprise = -log(parents_prediction_mean)
-        end
-
-        #If the input precision is finite
-    else
-        #Get the surprise
-        surprise =
-            -log(
-                parents_prediction_mean * pdf(
-                    Normal(
-                        node.parameters.category_means[1],
-                        node.parameters.input_precision,
-                    ),
-                    node.states.input_value,
-                ) +
-                (1 - parents_prediction_mean) * pdf(
-                    Normal(
-                        node.parameters.category_means[2],
-                        node.parameters.input_precision,
-                    ),
-                    node.states.input_value,
-                ),
-            )
+    #If a 1 was observed
+    if node.states.input_value == 0
+        #Get surprise
+        surprise = -log(1 - parents_prediction_mean)
+
+        #If a 0 was observed
+    elseif node.states.input_value == 1
+        #Get surprise
+        surprise = -log(parents_prediction_mean)
     end
 
     return surprise

test/testsuite/Aqua.jl

@@ -1,3 +1,3 @@
 using HierarchicalGaussianFiltering
 using Aqua
-Aqua.test_all(HierarchicalGaussianFiltering, ambiguities = false)
+Aqua.test_all(HierarchicalGaussianFiltering, ambiguities = false, persistent_tasks = false)

test/testsuite/test_canonical.jl

@@ -92,8 +92,6 @@ using Plots
 
         #Set parameters
         test_parameters = Dict(
-            ("u", "category_means") => [0.0, 1.0],
-            ("u", "input_precision") => Inf,
             ("xprob", "volatility") => -2.5,
             ("xvol", "volatility") => -6.0,
             ("xbin", "xprob", "coupling_strength") => 1.0,

test/testsuite/test_fit_model.jl

@@ -75,8 +75,6 @@ using Turing
 
         #Set fixed parameters and priors
         test_fixed_parameters = Dict(
-            ("u", "category_means") => Real[0.0, 1.0],
-            ("u", "input_precision") => Inf,
             ("xprob", "initial_mean") => 3.0,
             ("xprob", "initial_precision") => exp(2.306),
             ("xvol", "initial_mean") => 3.2189,