Add new features (dynamic exog coefs), new forecasting procedure, inp…

Project.toml

@@ -1,7 +1,7 @@
 name = "StateSpaceLearning"
 uuid = "971c4b7c-2c4e-4bac-8525-e842df3cde7b"
 authors = ["andreramosfc <[email protected]>"]
-version = "1.4.3"
+version = "2.0.0"
 
 [deps]
 Distributions = "31c24e10-a181-5473-b8eb-7969acd0382f"

README.md

@@ -20,35 +20,41 @@ model = StructuralModel(y)
 fit!(model)
 
 # Point Forecast
-prediction = StateSpaceLearning.forecast(model, 12) #Gets a 12 steps ahead prediction
+prediction = forecast(model, 12) # Gets a 12 steps ahead prediction
 
 # Scenarios Path Simulation
-simulation = StateSpaceLearning.simulate(model, 12, 1000) #Gets 1000 scenarios path of 12 steps ahead predictions
+simulation = simulate(model, 12, 1000) # Gets 1000 scenarios path of 12 steps ahead predictions
 ```
 
 ## StructuralModel Arguments
 
-* `y::Vector`: Vector of data.
-* `level::Bool`: Boolean where to consider intercept in the model (default: true)
-* `stochastic_level::Bool`: Boolean where to consider stochastic level component in the model (default: true)
-* `trend::Bool`: Boolean where to consider trend component in the model (default: true)
-* `stochastic_trend::Bool`: Boolean where to consider stochastic trend component in the model (default: true)
-* `seasonal::Bool`: Boolean where to consider seasonal component in the model (default: true)
-* `stochastic_seasonal::Bool`: Boolean where to consider stochastic seasonal component in the model (default: true)
-* `freq_seasonal::Int`: Seasonal frequency to be considered in the model (default: 12)
-* `outlier::Bool`: Boolean where to consider outlier component in the model (default: true)
-* `ζ_ω_threshold::Int`: Argument to stabilize `stochastic trend` and `stochastic seasonal` components (default: 12)
+* `y::Union{Vector,Matrix}`: Time series data
+* `level::String`: Level component type: "stochastic", "deterministic", or "none" (default: "stochastic")
+* `slope::String`: Slope component type: "stochastic", "deterministic", or "none" (default: "stochastic")
+* `seasonal::String`: Seasonal component type: "stochastic", "deterministic", or "none" (default: "stochastic")
+* `cycle::String`: Cycle component type: "stochastic", "deterministic", or "none" (default: "none")
+* `freq_seasonal::Union{Int,Vector{Int}}`: Seasonal frequency or vector of frequencies (default: 12)
+* `cycle_period::Union{Union{Int,<:AbstractFloat},Vector{Int},Vector{<:AbstractFloat}}`: Cycle period or vector of periods (default: 0)
+* `outlier::Bool`: Include outlier component (default: true)
+* `ζ_threshold::Int`: Threshold for slope innovations (default: 12)
+* `ω_threshold::Int`: Threshold for seasonal innovations (default: 12)
+* `ϕ_threshold::Int`: Threshold for cycle innovations (default: 12)
+* `stochastic_start::Int`: Starting point for stochastic components (default: 1)
+* `exog::Matrix`: Matrix of exogenous variables (default: zeros(length(y), 0))
+* `dynamic_exog_coefs::Union{Vector{<:Tuple}, Nothing}`: Dynamic exogenous coefficients (default: nothing)
 
 ## Features
 
 Current features include:
-* Estimation
-* Components decomposition
-* Forecasting
-* Completion of missing values
-* Predefined models, including:
-* Outlier detection
-* Outlier robust models
+* Model estimation using elastic net based regularization
+* Automatic component decomposition (trend, seasonal, cycle)
+* Point forecasting and scenario simulation
+* Missing value imputation
+* Outlier detection and robust modeling
+* Multiple seasonal frequencies support
+* Deterministic and stochastic component options
+* Dynamic exogenous variable handling
+* Best subset selection for exogenous variables
 
 ## Quick Examples
 
@@ -67,7 +73,7 @@ steps_ahead = 30
 
 model = StructuralModel(log_air_passengers)
 fit!(model)
-prediction_log = StateSpaceLearning.forecast(model, steps_ahead) # arguments are the output of the fitted model and number of steps ahead the user wants to forecast
+prediction_log = forecast(model, steps_ahead) # arguments are the output of the fitted model and number of steps ahead the user wants to forecast
 prediction = exp.(prediction_log)
 
 plot_point_forecast(airp.passengers, prediction)
@@ -76,7 +82,7 @@ plot_point_forecast(airp.passengers, prediction)
 
 ```julia
 N_scenarios = 1000
-simulation = StateSpaceLearning.simulate(model, steps_ahead, N_scenarios) # arguments are the output of the fitted model, number of steps ahead the user wants to forecast and number of scenario paths
+simulation = simulate(model, steps_ahead, N_scenarios) # arguments are the output of the fitted model, number of steps ahead the user wants to forecast and number of scenario paths
 
 plot_scenarios(airp.passengers, exp.(simulation))
 
@@ -97,22 +103,20 @@ log_air_passengers = log.(airp.passengers)
 model = StructuralModel(log_air_passengers)
 fit!(model)
 
-level = model.output.components["μ1"]["Values"] + model.output.components["ξ"]["Values"]
-slope = model.output.components["ν1"]["Values"] + model.output.components["ζ"]["Values"]
-seasonal = model.output.components["γ1_12"]["Values"] + model.output.components["ω_12"]["Values"]
-trend = level + slope
-
-plot(trend, w=2 , color = "Black", lab = "Trend Component", legend = :outerbottom)
-plot(seasonal, w=2 , color = "Black", lab = "Seasonal Component", legend = :outerbottom)
+# Access decomposed components directly
+trend = model.output.decomposition["trend"]
+seasonal = model.output.decomposition["seasonal_12"]
 
+plot(trend, w=2, color = "Black", lab = "Trend Component", legend = :outerbottom)
+plot(seasonal, w=2, color = "Black", lab = "Seasonal Component", legend = :outerbottom)
 ```
 
 | ![quick_example_trend](./docs/src/assets/trend.svg) | ![quick_example_seas](./docs/src/assets/seasonal.svg)|
 |:------------------------------:|:-----------------------------:|
 
 
-### Best Subset Selection
-Quick example on how to perform best subset selection in time series utilizing StateSpaceLearning.
+### Best Subset Selection and Dynamic Coefficients
+Example of performing best subset selection and using dynamic coefficients:
 
 ```julia
 using StateSpaceLearning
@@ -122,22 +126,33 @@ using Random
 
 Random.seed!(2024)
 
+# Load data
 airp = CSV.File(StateSpaceLearning.AIR_PASSENGERS) |> DataFrame
 log_air_passengers = log.(airp.passengers)
-X = rand(length(log_air_passengers), 10) # Create 10 exogenous features 
-β = rand(3)
-
-y = log_air_passengers + X[:, 1:3]*β # add to the log_air_passengers series a contribution from only 3 exogenous features.
-
-model = StructuralModel(y; Exogenous_X = X)
-fit!(model; α = 1.0, information_criteria = "bic", ϵ = 0.05, penalize_exogenous = true, penalize_initial_states = true)
-
-Selected_exogenous = model.output.components["Exogenous_X"]["Selected"]
 
+# Create exogenous features
+X = rand(length(log_air_passengers), 10)
+β = rand(3)
+y = log_air_passengers + X[:, 1:3]*β
+
+# Create model with exogenous variables
+model = StructuralModel(y; 
+    exog = X
+)
+
+# Fit model with elastic net regularization
+fit!(model; 
+    α = 1.0, # 1.0 for Lasso, 0.0 for Ridge
+    information_criteria = "bic",
+    ϵ = 0.05,
+    penalize_exogenous = true,
+    penalize_initial_states = true
+)
+
+# Get selected features
+selected_exog = model.output.components["exog"]["Selected"]
 ```
 
-In this example, the selected exogenous features were 1, 2, 3, as expected.
-
 ### Missing values imputation
 Quick example of completion of missing values for the air passengers time-series (artificial NaN values are added to the original time-series).
 
@@ -209,7 +224,7 @@ fit!(model)
 residuals_variances = model.output.residuals_variances
 
 ss_model = BasicStructural(log_air_passengers, 12)
-set_initial_hyperparameters!(ss_model, Dict("sigma2_ε" => residuals_variances["ε"], 
+StateSpaceModels.set_initial_hyperparameters!(ss_model, Dict("sigma2_ε" => residuals_variances["ε"], 
                                          "sigma2_ξ" =>residuals_variances["ξ"], 
                                          "sigma2_ζ" =>residuals_variances["ζ"], 
                                          "sigma2_ω" =>residuals_variances["ω_12"]))

docs/src/examples.md

@@ -46,7 +46,7 @@ y = solars[!, "y1"]
 T = length(y)
 steps_ahead = 48
 
-model = StructuralModel(y; freq_seasonal=24, trend=false, level=false)
+model = StructuralModel(y; freq_seasonal=24, slope="none", level="none", outlier=false)
 fit!(model; penalize_initial_states=false)
 simulation = StateSpaceLearning.simulate(model, steps_ahead, 100) #Gets a 12 steps ahead prediction
 plot_scenarios(y, simulation)
@@ -63,7 +63,7 @@ y = solars[!, "y1"]
 T = length(y)
 steps_ahead = 48
 
-model = StructuralModel(y; freq_seasonal=24, trend=false, level=false)
+model = StructuralModel(y; freq_seasonal=24, slope="none", level="none", outlier = false)
 fit!(model; penalize_initial_states=false)
 simulation = StateSpaceLearning.simulate(model, steps_ahead, 100; seasonal_innovation_simulation=24) #Gets a 12 steps ahead prediction
 plot_scenarios(y, simulation)
@@ -114,4 +114,35 @@ plot_point_forecast(y, prediction)
 ```
 ![two_seas](assets/two_seas.png)
 
-Note that the model was able to capture both seasonalities in this case.
+Note that the model was able to capture both seasonalities in this case.
+
+## Dynamic Exogenous Coefficients
+
+Dynamic exogenous coefficients allow the effect of exogenous variables to vary over time with specific patterns (e.g., level, slope, seasonal or cyclical). This is configured through the `dynamic_exog_coefs` parameter in the StructuralModel constructor.
+
+The `dynamic_exog_coefs` parameter accepts a vector of tuples, where each tuple contains:
+- First element: A vector of an exogenous variable
+- Second element: The name of the component that the exogenous variable will be associated with
+- Third element (optional): For the seasonal component, the freq_seasonal parameter and for cycle component, the cycle_period parameter.
+
+For example:
+```julia
+# Make X1's effect vary annually and X2's effect vary semi-annually
+airp = CSV.File(StateSpaceLearning.AIR_PASSENGERS) |> DataFrame
+y = log.(airp.passengers)
+X = vcat(collect(1:90), collect(90.5:-0.5:64)) + (rand(144) .* 10)
+y += X * -0.03
+
+dynamic_coefs = [(X, "level")]
+
+model = StructuralModel(y;
+    dynamic_exog_coefs=dynamic_coefs
+)
+
+fit!(model)
+
+prediction = forecast(model, 30; dynamic_exog_coefs_forecasts = [collect(63.5:-0.5:49)])
+
+plot_point_forecast(y, prediction)
+```
+![dynamic_exog](assets/dynamic_exog.png)

docs/src/features.md

@@ -104,10 +104,10 @@ X = rand(length(log_air_passengers), 10) # Create 10 exogenous features
 
 y = log_air_passengers + X[:, 1:3]*β # add to the log_air_passengers series a contribution from only 3 exogenous features.
 
-model = StructuralModel(y; Exogenous_X = X)
+model = StructuralModel(y; exog = X)
 fit!(model; α = 1.0, information_criteria = "bic", ϵ = 0.05, penalize_exogenous = true, penalize_initial_states = true)
 
-Selected_exogenous = model.output.components["Exogenous_X"]["Selected"]
+Selected_exogenous = model.output.components["exog"]["Selected"]
 
 ```
 

docs/src/manual.md

@@ -23,7 +23,14 @@ simulation = StateSpaceLearning.simulate(model, 12, 1000) #Gets 1000 scenarios p
 ```
 ## Models
 
-The package currently supports the implementation of the StructuralModel. If you have suggestions for additional models to include, we encourage you to contribute by opening an issue or submitting a pull request.
+The package currently supports the implementation of the StructuralModel, which includes capabilities for handling dynamic exogenous coefficients. If you have suggestions for additional models to include, we encourage you to contribute by opening an issue or submitting a pull request.
+
+```
+
+When using dynamic coefficients:
+- The model will create time-varying coefficients for each specified exogenous variable
+- Each coefficient will follow the specified cyclical pattern
+- When forecasting, you must provide future values for exogenous variables using the `Exogenous_Forecast` parameter
 
 ```@docs
 StateSpaceLearning.StructuralModel
@@ -39,7 +46,11 @@ StateSpaceLearning.fit!
 
 ## Forecasting and Simulating
 
-The package has functions to make point forecasts multiple steps ahead and to simulate scenarios based on those forecasts. These functions are implemented both for the univariate and to the multivariate cases.
+The package has functions to make point forecasts multiple steps ahead and to simulate scenarios based on those forecasts. These functions are implemented for both univariate and multivariate cases, with support for exogenous variables and dynamic coefficients.
+
+When using models with exogenous variables:
+- For standard exogenous variables, provide future values using the `Exogenous_Forecast` parameter
+- For dynamic coefficients, use the same `Exogenous_Forecast` parameter with values for each exogenous variable
 
 ```@docs
 StateSpaceLearning.forecast

paper_tests/m4_test/evaluate_model.jl

@@ -19,15 +19,13 @@ function evaluate_SSL(
 
     model = StateSpaceLearning.StructuralModel(
         normalized_y;
-        level=true,
-        stochastic_level=true,
-        trend=true,
-        stochastic_trend=true,
-        seasonal=true,
-        stochastic_seasonal=true,
+        level="stochastic",
+        slope="stochastic",
+        seasonal="stochastic",
         freq_seasonal=12,
         outlier=outlier,
-        ζ_ω_threshold=12,
+        ζ_threshold=12,
+        ω_threshold=12,
     )
     StateSpaceLearning.fit!(
         model;

paper_tests/m4_test/m4_test.jl

@@ -50,7 +50,7 @@ function run_config(
     save_init ? CSV.write(init_filepath, initialization_df) : nothing # Initialize empty CSV
 
     for i in 1:48000
-        if i in [10001, 20001, 30001, 40001] # Clear DataFrame to save memory
+        if i % 1000 == 1 # Clear DataFrame to save memory
             results_df = DataFrame()
             initialization_df = DataFrame()
         end
@@ -66,7 +66,8 @@ function run_config(
             information_criteria,
         )
 
-        if i in [10000, 20000, 30000, 40000, 48000]
+        if i % 1000 == 0
+            @info "Saving results for $i series"
             !save_init ? append_results(filepath, results_df) : nothing
             save_init ? append_results(init_filepath, initialization_df) : nothing
         end
@@ -100,9 +101,9 @@ end
 # Main script
 function main()
     results_table = DataFrame()
-    for outlier in [true]
-        for information_criteria in ["aic"]
-            for α in [0.1]
+    for outlier in [true, false]
+        for information_criteria in ["aic", "bic"]
+            for α in [0.0, 0.1, 0.3, 0.5, 0.7, 0.9, 1.0]
                 @info "Running SSL with outlier = $outlier, information_criteria = $information_criteria, α = $α"
                 results_table = run_config(
                     results_table, outlier, information_criteria, α, false, 60

paper_tests/m4_test/prepare_data.jl

@@ -2,7 +2,7 @@ function normalize(y::Vector, max_y::AbstractFloat, min_y::AbstractFloat)
     return (y .- min_y) ./ (max_y - min_y)
 end
 
-function de_normalize(y::Vector, max_y::AbstractFloat, min_y::AbstractFloat)
+function de_normalize(y, max_y::AbstractFloat, min_y::AbstractFloat)
     return (y .* (max_y - min_y)) .+ min_y
 end
 

paper_tests/simulation_test/evaluate_models.jl

@@ -28,7 +28,7 @@ function get_SSL_results(
         freq_seasonal=12,
         outlier=false,
         ζ_ω_threshold=12,
-        Exogenous_X=X_train,
+        exog=X_train,
     )
     t = @elapsed StateSpaceLearning.fit!(
         model;
@@ -39,13 +39,13 @@ function get_SSL_results(
         penalize_initial_states=true,
     )
 
-    selected = model.output.components["Exogenous_X"]["Selected"]
+    selected = model.output.components["exog"]["Selected"]
     true_positives, false_positives, false_negatives, true_negatives = get_confusion_matrix(
         selected, true_features, false_features
     )
 
-    mse = mse_func(model.output.components["Exogenous_X"]["Coefs"], true_β)
-    bias = bias_func(model.output.components["Exogenous_X"]["Coefs"], true_β)
+    mse = mse_func(model.output.components["exog"]["Coefs"], true_β)
+    bias = bias_func(model.output.components["exog"]["Coefs"], true_β)
 
     series_result = DataFrame(
         [

src/StateSpaceLearning.jl

@@ -1,6 +1,6 @@
 module StateSpaceLearning
 
-using LinearAlgebra, Statistics, GLMNet, Distributions, SparseArrays
+using LinearAlgebra, Statistics, GLMNet, Distributions, SparseArrays, Random
 
 abstract type StateSpaceLearningModel end
 
@@ -13,6 +13,12 @@ include("datasets.jl")
 include("fit_forecast.jl")
 include("plots.jl")
 
-export fit!, forecast, simulate, StructuralModel, plot_point_forecast, plot_scenarios
+export fit!,
+    forecast,
+    simulate,
+    StructuralModel,
+    plot_point_forecast,
+    plot_scenarios,
+    simulate_states
 
 end # module StateSpaceLearning