Checking derivatives.

Project.toml

@@ -43,6 +43,7 @@ Test = "8dfed614-e22c-5e08-85e1-65c5234f0b40"
 YAML = "ddb6d928-2868-570f-bddf-ab3f9cf99eb6"
 
 [weakdeps]
+Plots = "91a5bcdd-55d7-5caf-9e0b-520d859cae80"
 CUDA = "052768ef-5323-5732-b1bb-66c8b64840ba"
 oneAPI = "8f75cd03-7ff8-4ecb-9b8f-daf728133b1b"
 
@@ -53,3 +54,4 @@ DolooneAPIExt = "oneAPI"
 [compat]
 Dolang = "3.3.0"
 LabelledArrays = "≥1.16.0"
+Plots = "1.40.8"

demo.jl

@@ -1,10 +1,86 @@
-a = 4
-
 using Dolo
+using DoloYAML
+
+# model = include("examples/ymodels/consumption_savings.jl")
+model = include("examples/ymodels/rbc_iid.yaml")
+Dolo.time_iteration(model; improve=false)
+
+dmodel = Dolo.discretize(model)
+
+wk = Dolo.time_iteration_workspace(dmodel; improve=true)
+
+using FiniteDiff
+
+function residual_1(dmodel,wk,v)
+    x = Dolo.unravel(wk.x0, v)
+    r = Dolo.F(dmodel,x, wk.ψ)
+    Dolo.ravel(r)
+end
+
+ψ = deepcopy(wk.φ)
+
+function residual_2(dmodel,wk,v)
+    x = Dolo.unravel(wk.x0, v)
+    Dolo.fit!(ψ, x)
+    r = Dolo.F(dmodel,wk.x0, ψ)
+    Dolo.ravel(r)
+end
+
+
+
+v = Dolo.ravel(wk.x0)
+
+residual(dmodel, wk, v, wk.φ)
+
+Jdiff = FiniteDiff.finite_difference_jacobian(u->residual(dmodel, wk, u, ψ), v)
+
+J = Dolo.dF_1(dmodel, wk.x0, wk.φ)
+Jmat = convert(Matrix, J)
+
+@assert maximum( abs.(Jdiff - Jmat)./(1 .+ abs.(Jdiff)) ) <1e-6
+
+
+Jdiff_2 = FiniteDiff.finite_difference_jacobian(u->residual_2(dmodel, wk, u), v)
+J_2 = Dolo.dF_2(dmodel, wk.x0, wk.φ)
+
+Jmat_2 = convert(Matrix, J_2)
+
+using LinearMaps
+
+L_2 = convert(LinearMap, J_2)
+
+using Plots
+plot(xlims=(-10, 10), ylims=(-2,2))
+
+spy(Jmat)
+spy(Jmat_2, xlims=(1,150), ylims=(1,150))
+
+spy(Jmat_2)
+
+
+Δ = Jdiff_2 - Jmat_2
+maximum(abs.(Δ)./( 1 .+ Jmat_2))
+# # model = DoloYAML.yaml_import("examples/ymodels/consumption_savings.yaml")
+
+
+# wk = Dolo.time_iteration_workspace(dmodel; improve=true)
+
+# # @time Dolo.time_iteration(dmodel, wk; verbose=false, improve=false, improve_wait=10);
+
+# @time Dolo.time_iteration(dmodel, wk; 
+#     verbose=false, improve=false, improve_wait=10, engine=nothing);
+
+
+# dm.grid[1]
+
+# [dm.grid...]
+
 
+# # wk = Dolo.time_iteration_workspace(model; improve=true, interp_mode=:cubic)
+# # @time Dolo.time_iteration(model, wk; verbose=true, improve=true, improve_wait=20);
 
-model = include("examples/ymodels/consumption_savings.jl")
 
-Dolo.time_iteration(model)
 
+# # (r0, model, x0, φ, t_engine) = Dolo.time_iteration(model; verbose=false, improve=true);
 
+# # Dolo.F!(r0, model, x0, φ, t_engine)

examples/ymodels/consumption_savings_iid.jl

@@ -51,11 +51,11 @@ model = let
         (:w,:r,:e),
         SVector( [Q[i] for i=1:size(Q,1)]...  )
     )×CartesianSpace(;
-        y=[0.01, 100]
+        y=(0.01, 100.0)
     )
 
     controls = CartesianSpace(;
-        c=(0,Inf),
+        c=(0.0,Inf),
     )
     exogenous = Dolo.MarkovChain(
         (:w,:r,:e), P,Q

src/algos/time_iteration.jl

@@ -170,8 +170,9 @@ end
 function time_iteration(model::YModel; kwargs...)
     discr_options = get(kwargs, :discretization, Dict())
     interp_mode = get(kwargs, :interpolation, :cubic)
+    improve = get(kwargs, :improve, false)
     dmodel = discretize(model, discr_options)
-    wksp = time_iteration_workspace(dmodel; interp_mode=interp_mode)
+    wksp = time_iteration_workspace(dmodel; interp_mode=interp_mode, improve=improve)
     kwargs2 = pairs(NamedTuple( k=>v for (k,v) in kwargs if !(k in (:discretization, :interpolation))))
     time_iteration(dmodel, wksp; kwargs2...)
 end
@@ -202,7 +203,7 @@ function time_iteration(model::DYModel,
     # mem = typeof(workspace) <: Nothing ? time_iteration_workspace(model) : workspace
     mbsteps = 5
 
-    (;x0, x1, x2, dx, r0, J, φ) = workspace
+    (;x0, x1, x2, r0, dx, J, φ) = workspace
 
 
     local η_0 = NaN
@@ -235,6 +236,7 @@ function time_iteration(model::DYModel,
         trace && push!(ti_trace.data, deepcopy(φ))
 
         F!(r0, model, x0, φ, t_engine)
+
         # r0 = F(model, x0, φ)
 
         ε = norm(r0)
@@ -254,7 +256,7 @@ function time_iteration(model::DYModel,
         for k=1:max_bsteps
 
             F!(r0, model, x1,  φ, t_engine)
-            
+
             ε_n = norm(r0)
             if ε_n<tol_ε
                 iterations = t

src/dev_L2.jl

@@ -70,7 +70,8 @@ function *(L::LL, x0)
     Tf = getprecision(x0)
     r.data .*= convert(Tf, 0.0)
     mul!(r, L, x0)
-
+    r
+
 end
 
 # this takes 0.2 s !

src/funs.jl

@@ -30,7 +30,8 @@ function DFun(domain, values, itp)
     elseif eltype(values) <: SVector # && length(eltype(values)) == 1
         vars = (:y,)
     else
-        println(values)
+        # println(values)
+        nothing
     end
     return DFun(domain, values, itp, vars)
 end
@@ -53,7 +54,8 @@ function DFun(states, values::GVector{G,V}, vars=nothing; interp_mode=:linear) w
         elseif eltype(values) <: SVector # && length(eltype(values)) == 1
             vars = (:y,)
         else
-            println(values)
+            # println(values)
+            nothing
         end
     end
 
@@ -80,7 +82,8 @@ function DFun(states, values::GVector{G,V}, vars=nothing; interp_mode=:linear) w
         elseif eltype(values) <: SVector # && length(eltype(values)) == 1
             vars = (:y,)
         else
-            println(values)
+            # println(values)
+            nothing
         end
     end