Update stopping criteria of the solver

qntwrsm · May 13, 2024 · 3f51d81 · 3f51d81
1 parent 1ac5d51
commit 3f51d81
Show file tree

Hide file tree

Showing 2 changed files with 6 additions and 14 deletions.
diff --git a/src/fit/solver.jl b/src/fit/solver.jl
@@ -10,7 +10,7 @@ solver.jl
 =#
 
 """
-    update!(model)
+    update!(model, regularizer)
 
 Update parameters of dynamic factor model `model` using an
 expectation-maximization (EM) solve w/ regularization term given by
@@ -89,9 +89,7 @@ function update_loadings!(
 
         return (0.5 * dot(ΩΛ, Λ * Eff) - dot(ΩΛ, Eyf)) / length(V)
     end
-    ffb = FastForwardBackward(
-        stop=(iter, state) -> norm(state.res, Inf) < 1e-4
-    )
+    ffb = FastForwardBackward(maxit=1_000, tol=1e-4)
     (solution, _) = ffb(x0=loadings(F), f=objective, g=regularizer)
     loadings(F) .= solution
 
@@ -224,9 +222,7 @@ function update!(μ::Exogenous, y::AbstractMatrix, Σ::AbstractMatrix, regulariz
 
         return (0.5 * dot(Ωβ, β * XX) - dot(Ωβ, yX)) / size(regressors(μ), 2)
     end
-    ffb = FastForwardBackward(
-        stop=(iter, state) -> norm(state.res, Inf) < 1e-4
-    )
+    ffb = FastForwardBackward(maxit=1_000, tol=1e-4)
     (solution, _) = ffb(x0=slopes(μ), f=objective, g=regularizer)
     slopes(μ) .= solution
 
@@ -305,9 +301,7 @@ function update!(ε::SpatialAutoregression, Λ::AbstractMatrix, V::AbstractVecto
 
         return -logdet(G) + 0.5 * dot(Ω, Eee) / size(resid(ε), 2)
     end
-    ffb = FastForwardBackward(
-        stop=(iter, state) -> norm(state.res, Inf) < 1e-4
-    )
+    ffb = FastForwardBackward(maxit=1_000, tol=1e-4)
     (solution, _) = ffb(x0=logit.((spatial(ε) .+ offset) ./ scale), f=objective, g=regularizer)
     spatial(ε) .= scale .* logistic.(solution) .- offset
 
@@ -370,9 +364,7 @@ function update!(ε::SpatialMovingAverage, Λ::AbstractMatrix, V::AbstractVector
 
         return logdet(G) + 0.5 * tr(Σ \ Eee) / size(resid(ε), 2)
     end
-    ffb = FastForwardBackward(
-        stop=(iter, state) -> norm(state.res, Inf) < 1e-4
-    )
+    ffb = FastForwardBackward(maxit=1_000, tol=1e-4)
     (solution, _) = ffb(x0=logit.((spatial(ε) .+ offset) ./ scale), f=objective, g=regularizer)
     spatial(ε) .= scale .* logistic.(solution) .- offset
 

diff --git a/src/interface.jl b/src/interface.jl
@@ -200,7 +200,7 @@ function fit!(
     regularizer::NamedTuple=(factors=nothing, mean=nothing, error=nothing),
     init_method::NamedTuple=(factors=:data, mean=:data, error=:data), 
     ϵ::AbstractFloat=1e-4, 
-    max_iter::Integer=1000, 
+    max_iter::Integer=1_000, 
     verbose::Bool=false
 )
     keys(regularizer) ⊇ (:factors, :mean, :error) || error("regularizer must be a NamedTuple with keys :factors, :mean, and :error.")