1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
|
#
# Copyright 2007-2020 by the individuals mentioned in the source code history
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
library(OpenMx)
#Compare Nelder-Mead to the GD optimizer best at handling MxConstraints:
if(mxOption(NULL,"Default optimizer")!="SLSQP"){stop("SKIP")}
mxOption(key="feasibility tolerance", value = .001)
#The naive "soft" method only seems to work OK if EVERY vertex of the initial simplex is feasible...:
ism <- matrix(0.2,4,4) + diag(0.2,4)
colnames(ism) <- c("pred","pyellow","pgreen","pblue")
foo <- mxComputeNelderMead(iniSimplexMat=ism, nudgeZeroStarts=FALSE, xTolProx=1e-12, fTolProx=1e-8,eqConstraintMthd="soft")
#foo$verbose <- 5L
plan <- omxDefaultComputePlan()
plan$steps <- list(foo,plan$steps$RE)
#Run with SLSQP:
m1 <- mxModel(
"MultinomialWithLinearConstraints",
mxMatrix(type="Full",nrow=1,ncol=1,free=T,values=0.25,labels="pred",name="Pred",lbound=0,ubound=1),
mxMatrix(type="Full",nrow=1,ncol=1,free=T,values=0.25,labels="pyellow",name="Pyellow",lbound=0,ubound=1),
mxMatrix(type="Full",nrow=1,ncol=1,free=T,values=0.25,labels="pgreen",name="Pgreen",lbound=0,ubound=1),
mxMatrix(type="Full",nrow=1,ncol=1,free=T,values=0.25,labels="pblue",name="Pblue",lbound=0,ubound=1),
mxAlgebra( -2*(43*log(Pred) + 22*log(Pyellow) + 20*log(Pgreen) + 15*log(Pblue)), name="fitfunc"),
mxAlgebra( cbind(-2*43/Pred,-2*22/Pyellow,-2*20/Pgreen,-2*15/Pblue), name="objgrad",
dimnames=list(NULL,c("pred","pyellow","pgreen","pblue"))),
mxFitFunctionAlgebra(algebra="fitfunc",gradient="objgrad",numObs=100),
mxCI(c("pred","pyellow","pgreen","pblue")),
mxConstraint(Pred + Pyellow + Pgreen + Pblue - 1 == 0,name="indentifying")
)
m1run <- mxRun(m1)
summary(m1run)
m2 <- mxModel(
"MultinomialWithLinearConstraints",
plan,
mxMatrix(type="Full",nrow=1,ncol=1,free=T,values=0.25,labels="pred",name="Pred",lbound=0,ubound=1),
mxMatrix(type="Full",nrow=1,ncol=1,free=T,values=0.25,labels="pyellow",name="Pyellow",lbound=0,ubound=1),
mxMatrix(type="Full",nrow=1,ncol=1,free=T,values=0.25,labels="pgreen",name="Pgreen",lbound=0,ubound=1),
mxMatrix(type="Full",nrow=1,ncol=1,free=T,values=0.25,labels="pblue",name="Pblue",lbound=0,ubound=1),
mxAlgebra( -2*(43*log(Pred) + 22*log(Pyellow) + 20*log(Pgreen) + 15*log(Pblue)), name="fitfunc"),
mxAlgebra( cbind(-2*43/Pred,-2*22/Pyellow,-2*20/Pgreen,-2*15/Pblue), name="objgrad",
dimnames=list(NULL,c("pred","pyellow","pgreen","pblue"))),
mxFitFunctionAlgebra(algebra="fitfunc",gradient="objgrad",numObs=100),
mxCI(c("pred","pyellow","pgreen","pblue")),
mxConstraint(Pred + Pyellow + Pgreen + Pblue - 1 == 0,name="indentifying")
)
m2run <- mxRun(m2)
summary(m2run)
#The "soft" heuristic is now about as good as the default "GDsearch":
omxCheckCloseEnough(m1run$output$estimate, m2run$output$estimate, 1e-4)
#Run with Nelder-Mead, with different arguments:
ism3 <- ism
ism3[4,4] <- 0.46
#^^^Note that now, it is not the case that all of the initial vertices are feasible
colnames(ism3) <- c("pred","pyellow","pgreen","pblue")
foo3 <- mxComputeNelderMead(
iniSimplexMat=ism3, nudgeZeroStarts=FALSE, xTolProx=1e-12, fTolProx=1e-8, eqConstraintMthd="backtrack")
#foo3$verbose <- 5L
plan3 <- omxDefaultComputePlan()
plan3$steps <- list(foo3,plan3$steps$RE)
m3 <- mxModel(
"MultinomialWithLinearConstraints",
plan3,
mxMatrix(type="Full",nrow=1,ncol=1,free=T,values=0.25,labels="pred",name="Pred",lbound=0,ubound=1),
mxMatrix(type="Full",nrow=1,ncol=1,free=T,values=0.25,labels="pyellow",name="Pyellow",lbound=0,ubound=1),
mxMatrix(type="Full",nrow=1,ncol=1,free=T,values=0.25,labels="pgreen",name="Pgreen",lbound=0,ubound=1),
mxMatrix(type="Full",nrow=1,ncol=1,free=T,values=0.25,labels="pblue",name="Pblue",lbound=0,ubound=1),
mxAlgebra( -2*(43*log(Pred) + 22*log(Pyellow) + 20*log(Pgreen) + 15*log(Pblue)), name="fitfunc"),
mxAlgebra( cbind(-2*43/Pred,-2*22/Pyellow,-2*20/Pgreen,-2*15/Pblue), name="objgrad",
dimnames=list(NULL,c("pred","pyellow","pgreen","pblue"))),
mxFitFunctionAlgebra(algebra="fitfunc",gradient="objgrad",numObs=100),
mxCI(c("pred","pyellow","pgreen","pblue")),
mxConstraint(Pred + Pyellow + Pgreen + Pblue - 1 == 0,name="indentifying")
)
m3run <- mxRun(m3)
#The backtrack method isn't that helpful in this case:
summary(m3run)
#l1p:
foo4 <- mxComputeNelderMead(
iniSimplexMat=ism3, nudgeZeroStarts=FALSE, xTolProx=1e-12, fTolProx=1e-8, eqConstraintMthd="l1p")
#foo3$verbose <- 5L
plan4 <- omxDefaultComputePlan()
plan4$steps <- list(foo4,plan4$steps$RE)
m4 <- mxModel(
"MultinomialWithLinearConstraints",
plan4,
mxMatrix(type="Full",nrow=1,ncol=1,free=T,values=0.25,labels="pred",name="Pred",lbound=0,ubound=1),
mxMatrix(type="Full",nrow=1,ncol=1,free=T,values=0.25,labels="pyellow",name="Pyellow",lbound=0,ubound=1),
mxMatrix(type="Full",nrow=1,ncol=1,free=T,values=0.25,labels="pgreen",name="Pgreen",lbound=0,ubound=1),
mxMatrix(type="Full",nrow=1,ncol=1,free=T,values=0.25,labels="pblue",name="Pblue",lbound=0,ubound=1),
mxAlgebra( -2*(43*log(Pred) + 22*log(Pyellow) + 20*log(Pgreen) + 15*log(Pblue)), name="fitfunc"),
mxAlgebra( cbind(-2*43/Pred,-2*22/Pyellow,-2*20/Pgreen,-2*15/Pblue), name="objgrad",
dimnames=list(NULL,c("pred","pyellow","pgreen","pblue"))),
mxFitFunctionAlgebra(algebra="fitfunc",gradient="objgrad",numObs=100),
mxCI(c("pred","pyellow","pgreen","pblue")),
mxConstraint(Pred + Pyellow + Pgreen + Pblue - 1 == 0,name="indentifying")
)
m4run <- mxRun(m4)
#The l1p isn't that helpful in this case, either:
summary(m4run)
#Penalized fit should be slightly greater than raw fit:
omxCheckTrue(m4run$compute$steps[[1]]$output$penalizedFit > m4run$output$fit)
#GDsearch:
foo5 <- mxComputeNelderMead(
iniSimplexMat=ism3, nudgeZeroStarts=FALSE, xTolProx=1e-12, fTolProx=1e-8, eqConstraintMthd="GDsearch")
#foo3$verbose <- 5L
plan5 <- omxDefaultComputePlan()
plan5$steps <- list(foo5,plan5$steps$RE)
m5 <- mxModel(
"MultinomialWithLinearConstraints",
plan5,
mxMatrix(type="Full",nrow=1,ncol=1,free=T,values=0.25,labels="pred",name="Pred",lbound=0,ubound=1),
mxMatrix(type="Full",nrow=1,ncol=1,free=T,values=0.25,labels="pyellow",name="Pyellow",lbound=0,ubound=1),
mxMatrix(type="Full",nrow=1,ncol=1,free=T,values=0.25,labels="pgreen",name="Pgreen",lbound=0,ubound=1),
mxMatrix(type="Full",nrow=1,ncol=1,free=T,values=0.25,labels="pblue",name="Pblue",lbound=0,ubound=1),
mxAlgebra( -2*(43*log(Pred) + 22*log(Pyellow) + 20*log(Pgreen) + 15*log(Pblue)), name="fitfunc"),
mxAlgebra( cbind(-2*43/Pred,-2*22/Pyellow,-2*20/Pgreen,-2*15/Pblue), name="objgrad",
dimnames=list(NULL,c("pred","pyellow","pgreen","pblue"))),
mxFitFunctionAlgebra(algebra="fitfunc",gradient="objgrad",numObs=100),
mxCI(c("pred","pyellow","pgreen","pblue")),
mxConstraint(Pred + Pyellow + Pgreen + Pblue - 1 == 0,name="indentifying")
)
m5run <- mxRun(m5)
#The GDsearch method actually gets the answer:
summary(m5run)
omxCheckCloseEnough(m1run$output$estimate, m5run$output$estimate, 1e-4)
|
