Several bugfixes

Author: qntwrsm

src/DynamicFactorModels.jl

@@ -42,7 +42,7 @@ using Logging
 using StatsAPI: aic, aicc, bic
 
 import Statistics: mean, var, cov
-import StatsAPI: params, params!, fit!, loglikelihood, dof, nobs
+import StatsAPI: fit!, loglikelihood, dof, nobs
 import ProximalOperators: prox!
 
 export

src/fit/utilities.jl

@@ -194,7 +194,7 @@ function init!(ε::SpatialMovingAverage, method::Symbol, ξ::AbstractMatrix)
     return nothing
 end
 
-function params(model::DynamicFactorModel)
+function get_params(model::DynamicFactorModel)
     n = size(data(model), 1)
     R = nfactors(model)
 
@@ -210,12 +210,12 @@ function params(model::DynamicFactorModel)
 
     # parameters
     θ = zeros(n_params)
-    params!(θ, model)
+    get_params!(θ, model)
 
     return θ
 end
 
-function params!(θ::AbstractVector, model::DynamicFactorModel)
+function get_params!(θ::AbstractVector, model::DynamicFactorModel)
     idx = 1
 
     # loadings
@@ -284,7 +284,7 @@ function params!(θ::AbstractVector, model::DynamicFactorModel)
 
     return nothing
 end
-function params!(model::DynamicFactorModel, θ::AbstractVector)
+function set_params!(model::DynamicFactorModel, θ::AbstractVector)
     idx = 1
 
     # loadings
@@ -310,8 +310,7 @@ function params!(model::DynamicFactorModel, θ::AbstractVector)
         idx += offset
         # variance
         offset = length(cov(process(model)))
-        vec(cov(process(model)).mat) .= view(θ, idx:(idx + offset - 1))
-        cov(process(model)).chol.factors .= cholesky(Hermitian(cov(process(model)).mat)).factors
+        vec(cov(process(model)).data) .= view(θ, idx:(idx + offset - 1))
         idx += offset
     elseif process(model) isa UnrestrictedUnitRoot
         # variance
@@ -325,8 +324,7 @@ function params!(model::DynamicFactorModel, θ::AbstractVector)
         idx += offset
         # variance
         offset = length(cov(process(model)))
-        vec(cov(process(model)).mat) .= view(θ, idx:(idx + offset - 1))
-        cov(process(model)).chol.factors .= cholesky(Hermitian(cov(process(model)).mat)).factors
+        vec(cov(process(model)).data) .= view(θ, idx:(idx + offset - 1))
         idx += offset
     elseif process(model) isa NelsonSiegelUnitRoot
         # variance

src/interface.jl

@@ -370,15 +370,15 @@ function model_tuning_cv!(model::DynamicFactorModel, space::Dict, regularizer::F
 
         # fit on train sample
         fit!(train_model, regularizer = regularizer(params))
-        θ = params(train_model)
+        θ = get_params(train_model)
 
         # evaluate on test samples
         loss = zero(eltype(data(model)))
         for (t, test_model) in pairs(test_models)
             # out-of-sample data
             y = view(data(model), :, (Ttrain + t):(Ttrain + t + periods - 1))
             # loss
-            params!(test_model, θ)
+            set_params!(test_model, θ)
             loss += sum(loss_function, y - forecast(test_model, periods))
         end
 
@@ -415,7 +415,7 @@ function forecast(model::DynamicFactorModel, periods::Integer)
     # filter
     (a, _, _, _) = filter(model, predict = true)
     a_hat = similar(a[end])
-    forecasts = similar(data(model), size(model)[1], periods)
+    forecasts = similar(data(model), size(data(model), 1), periods)
 
     # forecast mean
     μ_hat = forecast(mean(model), periods)

src/utilities.jl

@@ -38,15 +38,15 @@ end
 Select a sample `sample` of the data from the dynamic factor process `F`.
 """
 function select_sample(F::UnrestrictedStationaryIdentified, sample::AbstractUnitRange)
-    UnrestrictedStationary((size(loadings(F), 1), length(sample), size(F)),
+    UnrestrictedStationary((size(loadings(F), 1), length(sample), nfactors(F)),
                            dependence = :identified, type = eltype(factors(F)))
 end
 function select_sample(F::UnrestrictedStationaryFull, sample::AbstractUnitRange)
-    UnrestrictedStationary((size(loadings(F), 1), length(sample), size(F)),
+    UnrestrictedStationary((size(loadings(F), 1), length(sample), nfactors(F)),
                            dependence = :full, type = eltype(factors(F)))
 end
 function select_sample(F::UnrestrictedUnitRoot, sample::AbstractUnitRange)
-    UnrestrictedUnitRoot((size(loadings(F), 1), length(sample), size(F)),
+    UnrestrictedUnitRoot((size(loadings(F), 1), length(sample), nfactors(F)),
                          type = eltype(factors(F)))
 end
 function select_sample(F::NelsonSiegelStationary, sample::AbstractUnitRange)
@@ -63,7 +63,7 @@ Simulate the dynamic factors from the dynamic factor process `F` `S` times using
 number generator `rng`.
 """
 function simulate(F::AbstractFactorProcess, S::Integer; rng::AbstractRNG = Xoshiro())
-    R = size(loadings(F), 2)
+    R = nfactors(F)
     f = similar(factors(F), R, S)
     dist = MvNormal(cov(F))
     for (s, fs) in pairs(eachcol(f))
@@ -186,7 +186,7 @@ function filter(model::DynamicFactorModel; predict::Bool = false)
     F = similar(y, typeof(P1))
 
     # initialize storage
-    ZtFinv = similar(P1)
+    ZtFinv = similar(Z')
     att = similar(a1)
     Ptt = similar(P1)