Merge pull request #112 from ilabcode/dev

Version 0.5.2

.github/workflows/CI_full.yml

@@ -37,7 +37,7 @@ jobs:
         with:
           version: ${{ matrix.version }}
           arch: ${{ matrix.arch }}
-      - uses: julia-actions/cache@v1
+      - uses: julia-actions/cache@v2
       - uses: julia-actions/julia-buildpkg@v1
       - uses: julia-actions/julia-runtest@v1
       - uses: julia-actions/julia-processcoverage@v1

.github/workflows/CI_small.yml

@@ -33,7 +33,7 @@ jobs:
         with:
           version: ${{ matrix.version }}
           arch: ${{ matrix.arch }}
-      - uses: julia-actions/cache@v1
+      - uses: julia-actions/cache@v2
       - uses: julia-actions/julia-buildpkg@v1
       - uses: julia-actions/julia-runtest@v1
       - uses: julia-actions/julia-processcoverage@v1

Project.toml

@@ -4,7 +4,7 @@ authors = ["Peter Thestrup Waade ptw@cas.au.dk",
             "Anna Hedvig Møller hedvig.2808@gmail.com",
             "Jacopo Comoglio jacopo.comoglio@gmail.com",
             "Christoph Mathys chmathys@cas.au.dk"]
-version = "0.5.1"
+version = "0.5.2"
 
 [deps]
 DataFrames = "a93c6f00-e57d-5684-b7b6-d8193f3e46c0"
@@ -16,9 +16,11 @@ Turing = "fce5fe82-541a-59a6-adf8-730c64b5f9a0"
 RecipesBase = "3cdcf5f2-1ef4-517c-9805-6587b60abb01"
 
 [compat]
-DataFrames = "1"
+DataFrames = "1.6"
 Distributions = "0.25"
 RecipesBase = "1"
-Turing = "0.30"
+Turing = "0.31"
 julia = "1.9"
 ReverseDiff = "1"
+Distributed = "1.10"
+Logging = "1.10"

README.md

@@ -3,8 +3,8 @@
 [![Stable](https://img.shields.io/badge/docs-stable-blue.svg)](https://ilabcode.github.io/ActionModels.jl)
 [![Build Status](https://github.com/ilabcode/ActionModels.jl/actions/workflows/CI_full.yml/badge.svg?branch=main)](https://github.com/ilabcode/ActionModels.jl/actions/workflows/CI_full.yml?query=branch%3Amain)
 [![Coverage](https://codecov.io/gh/ilabcode/ActionModels.jl/branch/main/graph/badge.svg?token=NVFiiPydFA)](https://codecov.io/gh/ilabcode/ActionModels.jl)
-[![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](https://opensource.org/licenses/MIT)
 [![License: GNU](https://img.shields.io/badge/License-GNU-yellow)](<https://www.gnu.org/licenses/>)
+[![Aqua QA](https://raw.githubusercontent.com/JuliaTesting/Aqua.jl/master/badge.svg)](https://github.com/JuliaTesting/Aqua.jl)
 
 ## Welcome to the ActionModels.jl package!
 

docs/Project.toml

@@ -1,5 +1,6 @@
 [deps]
 ActionModels = "320cf53b-cc3b-4b34-9a10-0ecb113566a3"
+Aqua = "4c88cf16-eb10-579e-8560-4a9242c79595"
 CSV = "336ed68f-0bac-5ca0-87d4-7b16caf5d00b"
 DataFrames = "a93c6f00-e57d-5684-b7b6-d8193f3e46c0"
 Distributions = "31c24e10-a181-5473-b8eb-7969acd0382f"

docs/src/Using_the_package/creating_own_action_model.jl

@@ -155,8 +155,7 @@ end
 
 #Set the parameters:
 
-parameters =
-    Dict("learning_rate" => 1, "action_precision" => 1, ("initial", "value") => 0)
+parameters = Dict("learning_rate" => 1, "action_precision" => 1, ("initial", "value") => 0)
 
 # We set the initial state parameter for "value" state because we need a starting value in the update step. 
 

docs/src/Using_the_package/premade_agents_and_models.jl

@@ -60,23 +60,15 @@ get_parameters(agent)
 
 set_parameters!(
     agent,
-    Dict(
-        "learning_rate" => 0.79,
-        "action_precision" => 0.60,
-        ("initial", "value") => 1,
-    ),
+    Dict("learning_rate" => 0.79, "action_precision" => 0.60, ("initial", "value") => 1),
 )
 
 # ## Defining an agent with custom parameter values
 
 # If you know which parameter values you wish to use when defining your agent, you can specify them in the beginning as a dict() with parameter name as a string followed by the value.
 agent_custom_parameters = premade_agent(
     "premade_binary_rescorla_wagner_softmax",
-    Dict(
-        "learning_rate" => 0.7,
-        "action_precision" => 0.8,
-        ("initial", "value") => 1,
-    ),
+    Dict("learning_rate" => 0.7, "action_precision" => 0.8, ("initial", "value") => 1),
 )
 
 #and we can retrieve the new parameters with the get_parameters() function

docs/src/Using_the_package/variations_of_util.jl

@@ -88,8 +88,7 @@ actions = give_inputs!(agent, inputs)
 
 # Set a prior for the parameter we wish to fit
 using Distributions
-priors =
-    Dict("action_precision" => Normal(1, 0.5), "learning_rate" => Normal(1, 0.1))
+priors = Dict("action_precision" => Normal(1, 0.5), "learning_rate" => Normal(1, 0.1))
 
 # Fit the model
 fitted_model = fit_model(agent, priors, inputs, actions)

docs/src/julia_src_files/creating_own_action_model.jl

@@ -50,7 +50,7 @@ function continuous_rescorla_wagner_gaussian(agent::Agent, input::Real)
 
     ## Read in parameters from the agent
     learning_rate = agent.parameters["learning_rate"]
-    action_noise  = agent.parameters["action_noise"]
+    action_noise = agent.parameters["action_noise"]
 
     ## Read in states with an initial value
     old_value = agent.states["value"]
@@ -67,32 +67,25 @@ function continuous_rescorla_wagner_gaussian(agent::Agent, input::Real)
     action_distribution = Distributions.Normal(new_value, action_noise)
 
     ##Update the states and save them to agent's history
-    update_states!(agent, Dict(
-        "value" => new_value,
-        "input" => input,
-    ))
+    update_states!(agent, Dict("value" => new_value, "input" => input))
 
     ## return the action distribution to sample actions from
     return action_distribution
 end
 
 
 #-- define parameters and states --#
-parameters = Dict(
-    "learning_rate" => 0.8,
-    "action_noise" => 1,
-    InitialStateParameter("value") => 0)
+parameters =
+    Dict("learning_rate" => 0.8, "action_noise" => 1, InitialStateParameter("value") => 0)
 
-states = Dict(
-    "value" => missing,
-    "input" => missing)
+states = Dict("value" => missing, "input" => missing)
 
 #-- create agent --#
 agent = init_agent(
     continuous_rescorla_wagner_gaussian,
     parameters = parameters,
-    states = states
-    )
+    states = states,
+)
 
 
 #-- define observations --#
@@ -136,12 +129,15 @@ function binary_rescorla_wagner_softmax(agent::Agent, input::Union{Bool,Integer}
     action_distribution = Distributions.Bernoulli(action_probability)
 
     #Update states
-    update_states!(agent, Dict(
-        "value" => new_value,
-        "value_probability" => 1 / (1 + exp(-new_value)),
-        "action_probability" => action_probability,
-        "input" => input,
-    ))
+    update_states!(
+        agent,
+        Dict(
+            "value" => new_value,
+            "value_probability" => 1 / (1 + exp(-new_value)),
+            "action_probability" => action_probability,
+            "input" => input,
+        ),
+    )
 
     return action_distribution
 end
@@ -156,9 +152,7 @@ end
 #Set the parameters:
 
 parameters =
-    Dict("learning_rate" => 1,
-     "action_precision" => 1, 
-     InitialStateParameter("value") => 0)
+    Dict("learning_rate" => 1, "action_precision" => 1, InitialStateParameter("value") => 0)
 
 # We set the initial state parameter for "value" state because we need a starting value in the update step. 
 
@@ -167,7 +161,7 @@ states = Dict(
     "value" => missing,
     "value_probability" => missing,
     "action_probability" => missing,
-    "input" => missing
+    "input" => missing,
 )
 
 # And lastly the action model:

docs/src/julia_src_files/premade_agents_and_models.jl

@@ -60,23 +60,15 @@ get_parameters(agent)
 
 set_parameters!(
     agent,
-    Dict(
-        "learning_rate" => 0.79,
-        "action_precision" => 0.60,
-        ("initial", "value") => 1,
-    ),
+    Dict("learning_rate" => 0.79, "action_precision" => 0.60, ("initial", "value") => 1),
 )
 
 # ## Defining an agent with custom parameter values
 
 # If you know which parameter values you wish to use when defining your agent, you can specify them in the beginning as a dict() with parameter name as a string followed by the value.
 agent_custom_parameters = premade_agent(
     "binary_rescorla_wagner_softmax",
-    Dict(
-        "learning_rate" => 0.7,
-        "action_precision" => 0.8,
-        ("initial", "value") => 1,
-    ),
+    Dict("learning_rate" => 0.7, "action_precision" => 0.8, ("initial", "value") => 1),
 )
 
 #and we can retrieve the new parameters with the get_parameters() function

docs/src/julia_src_files/variations_of_util.jl

@@ -88,8 +88,7 @@ actions = give_inputs!(agent, inputs)
 
 # Set a prior for the parameter we wish to fit
 using Distributions
-priors =
-    Dict("action_precision" => Normal(1, 0.5), "learning_rate" => Normal(1, 0.1))
+priors = Dict("action_precision" => Normal(1, 0.5), "learning_rate" => Normal(1, 0.1))
 
 # Fit the model
 fitted_model = fit_model(agent, priors, inputs, actions)

src/ActionModels.jl

@@ -9,14 +9,17 @@ export init_agent, premade_agent, warn_premade_defaults, multiple_actions, check
 export create_agent_model, fit_model
 export plot_parameter_distribution,
     plot_predictive_simulation, plot_trajectory, plot_trajectory!
-export get_history, get_states, get_parameters, set_parameters!, reset!, give_inputs!, single_input!
+export get_history,
+    get_states, get_parameters, set_parameters!, reset!, give_inputs!, single_input!
 export get_posteriors, update_states!, set_save_history!
 export InitialStateParameter, ParameterGroup
 
 #Load premade agents
 function __init__()
-    premade_agents["binary_rescorla_wagner_softmax"] = premade_binary_rescorla_wagner_softmax
-    premade_agents["continuous_rescorla_wagner_gaussian"] = premade_continuous_rescorla_wagner_gaussian
+    premade_agents["binary_rescorla_wagner_softmax"] =
+        premade_binary_rescorla_wagner_softmax
+    premade_agents["continuous_rescorla_wagner_gaussian"] =
+        premade_continuous_rescorla_wagner_gaussian
 end
 
 #Types for agents and errors

src/create_agent/init_agent.jl

@@ -83,7 +83,9 @@ function init_agent(
     action_model::Function;
     substruct::Any = nothing,
     parameters::Dict = Dict(),
-    parameter_groups::Union{ParameterGroup, Vector{ParameterGroup}} = Vector{ParameterGroup}(),
+    parameter_groups::Union{ParameterGroup,Vector{ParameterGroup}} = Vector{
+        ParameterGroup,
+    }(),
     states::Union{Dict,Vector} = Dict(),
     settings::Dict = Dict(),
     save_history::Bool = true,
@@ -159,7 +161,7 @@ function init_agent(
             value = parameter_group.value,
             grouped_parameters = parameter_group.parameters,
         )
-    
+
         #Set the parameters 
         set_parameters!(agent, parameter_group, parameter_group.value)
 

src/fitting/create_model.jl

@@ -14,6 +14,7 @@ Create a Turing model object used for fitting an ActionModels agent.
     multiple_actions::Bool,
     impute_missing_actions::Bool,
 )
+
     #Initialize dictionaries for storing sampled parameters
     multilevel_parameters = Dict()
     agent_parameters = Dict()
@@ -155,17 +156,25 @@ Create a Turing model object used for fitting an ActionModels agent.
 
                     #Sample the action from the probability distribution
                     actions[group][timestep] ~ action_distribution
-
+
+                    #Save the action to the agent in case it needs it in the future
+                    agent.states["action"] = actions[group][timestep]
+
 
                     #If there are multiple actions
                 else
+
+                    #Initialize empty vector for storing actions
+                    actions = []
+
                     #Go through each separate action
                     for (action_idx, single_distribution) in enumerate(action_distribution)
 
                         # If missing actions are to be imputed, we do not care if the action exists
                         if !impute_missing_actions
                             # If we should not impute missing actions, we need to check if the action exists
-                            @inbounds action_exists = !ismissing(actions[group][timestep, action_idx])
+                            @inbounds action_exists =
+                                !ismissing(actions[group][timestep, action_idx])
                             # if the action doesn't exist, we skip this timestep
                             if !action_exists
                                 continue
@@ -174,8 +183,13 @@ Create a Turing model object used for fitting an ActionModels agent.
 
                         #Sample the action from the probability distribution
                         @inbounds actions[group][timestep, action_idx] ~ single_distribution
-
+
+                        #Save the action
+                        push!(actions, actions[group][timestep, action_idx])
                     end
+
+                    #Save the action to the agent, for models that need previous action
+                    agent.states["action"] = actions
                 end
             end
         end
@@ -191,4 +205,4 @@ Create a Turing model object used for fitting an ActionModels agent.
             throw(e)
         end
     end
-end
+end

src/fitting/fit_model.jl

@@ -46,7 +46,11 @@ function fit_model(
     input_cols::Vector = [:input],
     action_cols::Vector = [:action],
     fixed_parameters::Dict = Dict(),
-    sampler::Turing.Inference.InferenceAlgorithm = NUTS(-1, 0.65; adtype=AutoReverseDiff(true)),
+    sampler::Turing.Inference.InferenceAlgorithm = NUTS(
+        -1,
+        0.65;
+        adtype = AutoReverseDiff(true),
+    ),
     n_cores::Integer = 1,
     n_iterations::Integer = 1000,
     n_chains::Integer = 2,
@@ -56,7 +60,7 @@ function fit_model(
     sampler_kwargs...,
 )
     ### SETUP ###
-   
+
     #Convert column names to symbols
     independent_group_cols = Symbol.(independent_group_cols)
     multilevel_group_cols = Symbol.(multilevel_group_cols)
@@ -205,6 +209,13 @@ function fit_model(
 
         #Load ActionModels, Turing and sister packages on worker processes
         @everywhere @eval using ActionModels, Turing
+        if @isdefined HierarchicalGaussianFiltering
+            @everywhere @eval using HierarchicalGaussianFiltering
+        end
+        if @isdefined ActiveInference
+            @everywhere @eval using ActiveInference
+        end
+
         #Broadcast necessary information to workers
         @everywhere agent = $agent
         @everywhere fit_info_all = $fit_info_all
@@ -323,7 +334,11 @@ function fit_model(
     inputs::Array,
     actions::Array;
     fixed_parameters::Dict = Dict(),
-    sampler::Turing.Inference.InferenceAlgorithm = NUTS(-1, 0.65; adtype=AutoReverseDiff(true)),
+    sampler::Turing.Inference.InferenceAlgorithm = NUTS(
+        -1,
+        0.65;
+        adtype = AutoReverseDiff(true),
+    ),
     n_cores::Integer = 1,
     n_iterations::Integer = 1000,
     n_chains = 2,
@@ -334,8 +349,16 @@ function fit_model(
 )
 
     #Create column names
-    input_cols = map(x -> "input$x", 1:size(inputs, 2))
-    action_cols = map(x -> "action$x", 1:size(actions, 2))
+    input_cols = map(x -> "input$x", 1:length(first(inputs)))
+    action_cols = map(x -> "action$x", 1:length(first(actions)))
+
+    #Make vectors of vectors into arrays
+    if length(input_cols) > 1
+        inputs = mapreduce(permutedims, vcat, inputs)
+    end
+    if length(action_cols) > 1
+        actions = mapreduce(permutedims, vcat, actions)
+    end
 
     #Create dataframe of the inputs and actions
     data = DataFrame(hcat(inputs, actions), vcat(input_cols, action_cols))