File: MxRestore.R

package info (click to toggle)
r-cran-openmx 2.21.1%2Bdfsg-1
  • links: PTS, VCS
  • area: main
  • in suites: bookworm
  • size: 14,412 kB
  • sloc: cpp: 36,577; ansic: 13,811; fortran: 2,001; sh: 1,440; python: 350; perl: 21; makefile: 5
file content (257 lines) | stat: -rw-r--r-- 10,153 bytes parent folder | download | duplicates (2)
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
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
#
#   Copyright 2007-2021 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.

removeTrailingSeparator <- function(x) {
	return(sub('/$', '', x))
}

#' Save model state to a checkpoint file
#'
#' @template args-model
#' @template args-chkpt
#'
#' @description
#' The function saves the last state of a model to a checkpoint file.
#' 
#' @details
#' In general, the arguments \sQuote{chkpt.directory} and \sQuote{chkpt.prefix} should be identical to the \code{\link{mxOption}}: \sQuote{Checkpoint Directory} and \sQuote{Checkpoint Prefix} that were specified on the model before execution.
#' 
#' Alternatively, the checkpoint file can be manually loaded as a data.frame in R.  Use \code{\link{read.table}} with the options \code{header=TRUE, sep="\t", stringsAsFactors=FALSE, check.names=FALSE}.
#'
#' @return
#' Returns a logical indicating the success of writing the checkpoint file to the checkpoint directory.
#' @template ref-manual
#' @family model state
#' @examples
#' library(OpenMx)
#' 
#' # Simulate some data
#' 
#' x=rnorm(1000, mean=0, sd=1)
#' y= 0.5*x + rnorm(1000, mean=0, sd=1)
#' tmpFrame <- data.frame(x, y)
#' tmpNames <- names(tmpFrame)
#' 
#' dir <- tempdir()  # safe place to create files
#' mxOption(key="Checkpoint Directory", value=dir)
#'
#' # Create a model that includes an expected covariance matrix,
#' # an expectation function, a fit function, and an observed covariance matrix
#' 
#' data <- mxData(cov(tmpFrame), type="cov", numObs = 1000)
#' expCov <- mxMatrix(type="Symm", nrow=2, ncol=2, values=c(.2,.1,.2), free=TRUE, name="expCov")
#' expFunction <- mxExpectationNormal(covariance="expCov", dimnames=tmpNames)
#' fitFunction <- mxFitFunctionML()
#' testModel <- mxModel(model="testModel", expCov, data, expFunction, fitFunction)
#' 
#' #Use mxRun to optimize the free parameters in the expected covariance matrix
#' modelOut <- mxRun(testModel)
#' modelOut$expCov
#' 
#' # Save the ending state of modelOut in a checkpoint file
#' mxSave(modelOut)
#' 
#' # Restore the saved model from the checkpoint file
#' modelSaved <- mxRestore(testModel)
#' modelSaved$expCov
mxSave <- function(model, chkpt.directory = ".", chkpt.prefix = "") {
	if (!is(model, "MxModel")) {
		stop("'model' argument must be a MxModel object")
	}
	if (!missing(chkpt.directory)) model <- mxOption(model,"Checkpoint Directory", chkpt.directory)
	if (!missing(chkpt.prefix))    model <- mxOption(model,"Checkpoint Prefix", chkpt.prefix)
	model <- mxOption(model,"Checkpoint Units",'evaluations')
	model <- mxOption(model,"Checkpoint Count",1)
	model <- mxModel(model, mxComputeOnce('fitfunction', 'fit'))
	mxRun(model, checkpoint=TRUE, silent=TRUE)
	invisible(TRUE)
}

#' Restore model state from a checkpoint file
#'
#' @template args-model
#' @template args-chkpt
#' @param line integer. Which line from the checkpoint file to restore (defaults to the last line)
#' @param strict logical. Require that the checkpoint name and model name match
#'
#' @details
#' In general, the arguments \sQuote{chkpt.directory} and \sQuote{chkpt.prefix} should be identical to the \code{\link{mxOption}}: \sQuote{Checkpoint Directory} and \sQuote{Checkpoint Prefix} that were specified on the model before execution.
#'
#' Alternatively, the checkpoint file can be manually loaded as a data.frame in R and passed to \code{\link{mxRestoreFromDataFrame}}.
#' Use \code{\link{read.table}} with the options \code{header=TRUE, sep="\t", stringsAsFactors=FALSE, check.names=FALSE}.
#'
#' @return
#' Returns an MxModel object with free parameters updated to the last
#' saved values. When \sQuote{line} is provided, the MxModel is updated
#' to the values on that line within the checkpoint file.
#'
#' @template ref-manual
#' @family model state
#' @examples
#' library(OpenMx)
#' 
#' # Simulate some data
#' 
#' x=rnorm(1000, mean=0, sd=1)
#' y= 0.5*x + rnorm(1000, mean=0, sd=1)
#' tmpFrame <- data.frame(x, y)
#' tmpNames <- names(tmpFrame)
#' 
#' dir <- tempdir()  # safe place to create files
#' mxOption(key="Checkpoint Directory", value=dir)
#' 
#' # Create a model that includes an expected covariance matrix,
#' # an expectation function, a fit function, and an observed covariance matrix
#' 
#' data <- mxData(cov(tmpFrame), type="cov", numObs = 1000)
#' expCov <- mxMatrix(type="Symm", nrow=2, ncol=2, values=c(.2,.1,.2), free=TRUE, name="expCov")
#' expFunction <- mxExpectationNormal(covariance="expCov", dimnames=tmpNames)
#' fitFunction <- mxFitFunctionML()
#' testModel <- mxModel(model="testModel", expCov, data, expFunction, fitFunction)
#' 
#' #Use mxRun to optimize the free parameters in the expected covariance matrix
#' modelOut <- mxRun(testModel, checkpoint = TRUE)
#' modelOut$expCov
#' 
#' #Use mxRestore to load the last checkpoint saved state of the model
#' modelRestore <- mxRestore(testModel)
#' modelRestore$expCov

mxRestore <- function(model, chkpt.directory = mxOption(model, "Checkpoint directory"),
		      chkpt.prefix = mxOption(model,"Checkpoint Prefix"), line=NULL, strict=FALSE)
{
  warnModelCreatedByOldVersion(model)
	chkpt.directory <- removeTrailingSeparator(chkpt.directory)
	if (strict && chkpt.prefix == "") {
		chkpt.prefix <- model$name
	}
	pattern <- paste("^\\Q", chkpt.prefix, "\\E.*(\\.omx)$", sep = '')
	chkpt.files <- list.files(chkpt.directory, full.names = FALSE)
	chkpt.files <- grep(pattern, chkpt.files, perl=TRUE, value=TRUE)
	if (strict) {
		if (length(chkpt.files) > 1) {
			stop(paste("chkpt.prefix", omxQuotes(chkpt.prefix),
				   "matched more than one file"))
		} else if (length(chkpt.files) == 0) {
			stop(paste("Cannot find", omxQuotes(paste(model$name, 'omx', sep=".")),
				   "in", chkpt.directory))
		}
	} else {
		if(length(chkpt.files) == 0) {
			return(model)
		}
		# Move the most likely match to the end so those estimates take precedence.
		matchIndex <- match(paste(model$name, 'omx', sep="."), chkpt.files)
		if (!is.na(matchIndex)) {
			chkpt.files <- c(chkpt.files[-matchIndex], paste(model$name, 'omx', sep="."))
		}
	}
	if (length(chkpt.files) > 1 && !is.null(line)) {
		stop(paste("Ambiguous: cannot specify line =", line,
			   "with more than one checkpoint found:",
			   omxQuotes(chkpt.files)))
	}
	if (length(chkpt.files) > 1) {
		message(paste("Loading estimates from more than one checkpoint:",
			      omxQuotes(chkpt.files)))
	}
	for(i in 1:length(chkpt.files)) {
		filename <- chkpt.files[[i]]
		filepath <- paste(chkpt.directory, filename, sep = '/')
		checkpoint <- read.table(filepath, header=TRUE, stringsAsFactors=FALSE, check.names=FALSE, sep="\t")
		model <- mxRestoreFromDataFrame(model, checkpoint, line)
	}
	return(model)
}

#' @rdname mxRestore
#' @param checkpoint a data.frame containing the model state
mxRestoreFromDataFrame <- function(model, checkpoint, line=NULL)
{
  warnModelCreatedByOldVersion(model)
  allPar <- names(omxGetParameters(model, indep=TRUE, free=NA))
  badPar <- intersect(c("OpenMxContext","OpenMxNumFree","OpenMxEvals",
			"iterations","timestamp","objective"), allPar)
  if (any(badPar)) {
    warning(paste("You model", omxQuotes(model$name), "has illegal parameters",
		  "names:", omxQuotes(badPar)))
  }
  pick <- match(allPar, colnames(checkpoint))
  if (all(is.na(pick))) return(model)
  allVars <- !any(is.na(pick))
  if (is.null(line)) {
    row <- nrow(checkpoint)
  } else {
    row <- line
    if (row < 2 || row > nrow(checkpoint)) {
      warning(paste("Requested line", line,
		    "but checkpoint contains lines 2 to",
		    nrow(checkpoint), "; using the last line"))
      row <- nrow(checkpoint)
    }
  }
  pick <- pick[!is.na(pick)]
  varNames <- colnames(checkpoint)[pick]
  model <- omxSetParameters(model, labels=varNames, values=as.numeric(checkpoint[row, varNames]))
  if (allVars) {
    model@output$fit <- checkpoint[row, 'objective']
    model@output$fitUnits <- checkpoint[row, 'fitUnits']
    if (!is.null(model@output$fitUnits) && model@output$fitUnits == '-2lnL') {
      model@output$Minus2LogLikelihood <- model@output$fit
    }
    model@output$estimates <- coef(model)
    se <- paste0(varNames, "SE")
    if (all(!is.na(match(se, colnames(checkpoint))))) {
      model@output$standardErrors <- matrix(as.numeric(checkpoint[row, se]), ncol=1,
					    dimnames=list(varNames,NULL))
    }
    grad <- paste0(varNames, "Grad")
    if (all(!is.na(match(grad, colnames(checkpoint))))) {
      vec <- as.numeric(checkpoint[row, grad])
      names(vec) <- varNames
      model@output$gradient <- vec
    }
    vc <- rep('', length(varNames) * (length(varNames)+1) / 2)
    vx <- 1
    for (cx in 1:length(varNames)) {
      for (rx in cx:length(varNames)) {
	vc[vx] <- paste0('V', varNames[rx], ':', varNames[cx])
	vx <- vx + 1
      }
    }
    if (all(!is.na(match(vc, colnames(checkpoint))))) {
      vcov <- matrix(0, length(varNames), length(varNames),
		     dimnames=list(varNames, varNames))
      vcov[lower.tri(vcov, TRUE)] <- as.numeric(checkpoint[row, vc])
      vcov[upper.tri(vcov)] <- t(vcov)[upper.tri(vcov)]
      model@output$vcov <- vcov
    }

    # refer to MxRun.R
    unsafe <- TRUE
    namespace <- imxGenerateNamespace(model)
    flatModel <- imxFlattenModel(model, namespace, unsafe)
    dependencies <- cycleDetection(flatModel)
    dependencies <- transitiveClosure(flatModel, dependencies)
    freeVarGroups <- buildFreeVarGroupList(flatModel)
    flatModel <- generateParameterList(flatModel, dependencies, freeVarGroups)
    matrices <- generateMatrixList(flatModel)
    parameters <- flatModel@parameters

    model@.wasRun <- TRUE
    model@.modifiedSinceRun <- FALSE #?
  }
  model
}