separate function for translating optimization variables to variance …

…component parameters
OpenMendel · Jul 14, 2016 · 61d3140 · 61d3140
1 parent d709a3d
commit 61d3140
Showing 1 changed file with 26 additions and 31 deletions.
diff --git a/src/two_variance_component.jl b/src/two_variance_component.jl
@@ -497,6 +497,26 @@ function TwoVarCompOptProb{T}(
  TwoVarCompOptProb{T}(vcm, vcobsrot, qpsolver, L, ∇Σ, HΣ, HL, Xwork, ywork, Q, c)
 end
 
+"""
+Translate optimization variables to variance component parmeters `dd.vcmodel.Σ`.
+"""
+function optimparam_to_vcparam!{T}(dd::TwoVarCompOptProb, x::Vector{T})
+
+ d = size(dd.L[1], 1)
+ nvar = nvarparams(dd.vcmodel)
+ nvarhalf = div(nvar, 2)
+ # variance parameter
+ dd.L[1][trilind(d)] = x[1:nvarhalf]
+ dd.L[2][trilind(d)] = x[(nvarhalf+1):end]
+ # exponentiate diagonal entries
+ @inbounds @simd for i in 1:d
+ dd.L[1][i, i] = exp(dd.L[1][i, i])
+ dd.L[2][i, i] = exp(dd.L[2][i, i])
+ end
+ A_mul_Bt!(dd.vcmodel.Σ[1], dd.L[1], dd.L[1])
+ A_mul_Bt!(dd.vcmodel.Σ[2], dd.L[2], dd.L[2])
+end
+
 function MathProgBase.initialize(dd::TwoVarCompOptProb,
  requested_features::Vector{Symbol})
  for feat in requested_features
@@ -513,19 +533,8 @@ MathProgBase.eval_jac_g(dd::TwoVarCompOptProb, J, x) = nothing
 
 function MathProgBase.eval_f{T}(dd::TwoVarCompOptProb, x::Vector{T})
 
- d = size(dd.L[1], 1)
- nvar = nvarparams(dd.vcmodel)
- nvarhalf = div(nvar, 2)
- # variance parameter
- dd.L[1][trilind(d)] = x[1:nvarhalf]
- dd.L[2][trilind(d)] = x[(nvarhalf+1):end]
- # exponentiate diagonal entries
- @inbounds @simd for i in 1:d
- dd.L[1][i, i] = exp(dd.L[1][i, i])
- dd.L[2][i, i] = exp(dd.L[2][i, i])
- end
- A_mul_Bt!(dd.vcmodel.Σ[1], dd.L[1], dd.L[1])
- A_mul_Bt!(dd.vcmodel.Σ[2], dd.L[2], dd.L[2])
+ # update variance parameter from optim variable x
+ optimparam_to_vcparam!(dd, x)
  # update mean parameters
  update_meanparam!(dd.vcmodel, dd.vcdatarot,
  dd.qpsolver, dd.Xwork, dd.ywork, dd.Q, dd.c)
@@ -542,15 +551,8 @@ function MathProgBase.eval_grad_f{T}(
  d = size(dd.L[1], 1)
  nvar = nvarparams(dd.vcmodel)
  nvarhalf = div(nvar, 2)
- # variance parameter
- dd.L[1][trilind(d)] = x[1:nvarhalf]
- dd.L[2][trilind(d)] = x[(nvarhalf+1):end]
- @inbounds @simd for i in 1:d
- dd.L[1][i, i] = exp(dd.L[1][i, i])
- dd.L[2][i, i] = exp(dd.L[2][i, i])
- end
- A_mul_Bt!(dd.vcmodel.Σ[1], dd.L[1], dd.L[1])
- A_mul_Bt!(dd.vcmodel.Σ[2], dd.L[2], dd.L[2])
+ # update variance parameter from optim variable x
+ optimparam_to_vcparam!(dd, x)
  # update mean parameters
  update_meanparam!(dd.vcmodel, dd.vcdatarot,
  dd.qpsolver, dd.Xwork, dd.ywork, dd.Q, dd.c)
@@ -585,15 +587,8 @@ function MathProgBase.eval_hesslag{T}(dd::TwoVarCompOptProb, H::Vector{T},
  d = size(dd.L[1], 1)
  nvar = nvarparams(dd.vcmodel)
  nvarhalf = div(nvar, 2)
- # variance parameter
- dd.L[1][trilind(d)] = x[1:nvarhalf]
- dd.L[2][trilind(d)] = x[(nvarhalf+1):end]
- @inbounds @simd for i in 1:d
- dd.L[1][i, i] = exp(dd.L[1][i, i])
- dd.L[2][i, i] = exp(dd.L[2][i, i])
- end
- A_mul_Bt!(dd.vcmodel.Σ[1], dd.L[1], dd.L[1])
- A_mul_Bt!(dd.vcmodel.Σ[2], dd.L[2], dd.L[2])
+ # update variance parameter from optim variable x
+ optimparam_to_vcparam!(dd, x)
  # update mean parameters
  update_meanparam!(dd.vcmodel, dd.vcdatarot,
  dd.qpsolver, dd.Xwork, dd.ywork, dd.Q, dd.c)