Changes in OpenMx

Release 1.2.0-1926 (February 04, 2012)

• new interface such that 'objective@info$expMean‘, 'objective@info$expCov‘ and 'objective@info$likelihoods‘ are available in FIML or RAM objective functions.
• bug fix for omxSetParameters with lbound or ubound of NA.
• catching unknown matrix operator or function in mxAlgebra() or mxConstraint() declaration
• deprecated ‘all’ argument from mxPath function and replaced it with ‘connect’ argument. Also updated demos that used ‘all’ argument and documentation.
• removed ‘excludeself’ argument from mxPath() function
• added deprecation warning for argument ‘all’ = TRUE in mxPath()
• added support for OpenMP in hessian calculation
• added support for OpenMP in continuous FIML objective function
• added support for OpenMP in ordinal FIML objective function
• added support for OpenMP in joint ordinal/continuous FIML objective function
• added more descriptive error message to mxOption()
• serializing the sum operation on likelihoods - floating point addition is not associative

Release 1.1.2-1818 (October 24, 2011)

• fixed bug in summary() function, see http://openmx.psyc.virginia.edu/thread/1104
• fixed bug in independence model likelihood calculation for ML objective
• included support for gcc 4.5 and 4.6 under 32-bit x86

Release 1.1.1-1784 (September 11, 2011)

• fixed several bugs in joint ordinal-continuous integration
• fixed several typos in User Guide

Release 1.1.0-1764 (August 22, 2011)

• Joint and Ordinal documentation included and up to date.
• added deprecation warning for argument ‘all’=TRUE in mxPath()
• added omxSelectRowsAndCols, omxSelectRows and omxSelectCols documentation
• fixed model flattening to keep track of confidence intervals in submodels
• added error checking for dimnames on MxMatrix and MxAlgebra objects
• reformated comments and heading style for all demos
• fixed segmentation fault on backend error condition
• turn off jiggling of free parameters with starting values of 0.0 when useOptimizer=FALSE
• allow non-RAM objective functions in RAM model
• change model type names to ‘default’ and ‘RAM’
• no longer explicitly transforming RAM + raw data models into FIML models
• fixed bug in MxMatrix indexing operator
• added argument ‘threshnames’ to mxFIMLObjective() and mxRAMObjective()
• renaming majority of omx* functions to imx* functions. See http://openmx.psyc.virginia.edu/thread/761
• added error checking to mxOption() function
• added “Optimality tolerance” to mxOption() selection
• added ‘lbound’ and ‘ubound’ columns to summary() output of free parameters
• added asterisks to the ‘lbound’ and ‘ubound’ columns when feasibility tolerance is met
• added “omxNot” function to the set of available mxAlgebra() function
• added “omxSelectRows”, “omxSelectCols”, and “omxSelectRowsAndCols” as mxAlgebra operators()
• added “mean” function to the set of available mxAlgebra() functions
• added slots “expCov” and “expMean” to the MxRAMObjective function
• added useOptimizer option to mxRun.
• added error checking in frontend and backend for non-positive-definite observed covariance matrices
• added “omxGreaterThan”, “omxLessThan”, “omxApproxEquals”, “omxAnd”, and “omxOr” operators to the set of mxAlgebra() operators
• error checking for model[[1]] or model[[TRUE]]
• error checking in the front-end whether more than 20 ordinal columns are present in a data set
• improved performance in the front-end in mxModel() for adding paths to RAM models
• print name of algebra when operator has too few or too many arguments
• added mxErrorPool() function and R documentation.
• added Apache license information to all R documentation files.
• new implementation of mxEval().
• new argument ‘defvar.row’ to mxEval(). See ?mxEval.
• handling definition variables for (I - A) ^ - 1 speedup
• handling square bracket labels for (I - A) ^ - 1 speedup
• added argument ‘free’ to omxGetParameters. See ?omxGetParameters.
• added argument ‘strict’ to omxSetParameters. See ?omxSetParameters.
• eliminated warnings for confidence interval optimization codes
• added ”...” argument to mxRObjectiveFunction()
• fix memory leak in RAM objective function
• removed dependency to MBESS library in R documentation
• added more descriptive error message when thresholds are not sorted
• incorporated NaN unsafe matrix-matrix multiplication (dgemm) from R <= 2.11.1
• incorporated NaN unsafe matrix-vector multiplication (dgemv) from R <= 2.11.1
• return NA in mxVersion() if “OpenMx” cannot be found
• fix infinite loop in objective function transformations
• added initialization to load OpenMx on swift workers
• implemented omxParallelCI() to calculate confidence intervals in parallel
• only calculating CIs for upper triangle of symmetric matrices
• cleanup appearance of transient MxMatrix objects in error messages
• fix bug with very large number of omxUntitledName() objects
• added optional argument CPUS=n to “make test” target
• change snowfall interface to use sfClusterApplyLB()
• storing raw data in row-major order, and copying contiguous data rows
• fix bug in mxModel() when using remove = TRUE

Release 1.0.7-1706 (July 6, 2011)

• error checking in front-end for non-positive-definite observed covariance matrices
• fix bug in MxMatrix indexing operator
• added deprecation warning for argument ‘all’=TRUE in mxPath()

Release 1.0.6-1581 (March 10, 2011)

• Bug fix corner case in sorting data with definition variables

Release 1.0.5-1575 (March 8, 2011)

• added error checking in ML objective to match expected covariance and observed covariance matrices
• updated BootstrapParallel.R demo to use mxData() instead of model@data
• added the dataset from the Psychometrika article (www.springerlink.com/content/dg37445107026711)

Release 1.0.4-1540 (January 16, 2011)

• added initialization to load OpenMx on swift workers
• only calculating CIs for upper triangle on symmetric matrices
• fix bug with very large number of omxUntitledName() objects
• incorporated NaN unsafe matrix-vector multiplication (dgemv) from R <= 2.11.1
• fix bug in mxModel() when using remove = TRUE
• Added intervals to MxModel class documentation

Release 1.0.3-1505 (November 10, 2010)

• return NA in mxVersion() if “OpenMx” cannot be found
• eliminate infinite loop in objective function transformations

Release 1.0.2-1497 (November 5, 2010)

• missing <errno.h> include
• fix memory leak in RAM objective function
• incorporated NaN unsafe matrix-matrix multiplication (dgemm) from R <= 2.11.1

Release 1.0.1-1464 (October 8, 2010)

• bugfix for mxEval() and MxData objects
• handling definition variables for (I - A) ^ - 1 speedup
• handling square bracket labels for (I - A) ^ - 1 speedup
• added argument ‘free’ to omxGetParameters. See ?omxGetParameters.
• added argument ‘strict’ to omxSetParameters. See ?omxSetParameters.
• eliminated warnings for confidence interval optimization codes

Release 1.0.0-1448 (September 30, 2010)

• added missing entries to demo 00INDEX file

Release 0.9.2-1446 (September 26, 2010)

• added growth mixture models to user guide
• added initial Swift hook in omxLapply() - currenly activated only for mxRun calls
• fixed a bug in mxRename() when encountering symbol of missingness
• bugfix for crash when ‘*’ is used instead of ‘%*%’
• feature removal: square brackets in MxMatrix labels now accept only literal values
• bugfix for definition variables used in mxRowObjective (which is still experimental)
• bugfix for (I - A) ^ -1 speedup with FIML optimization

Release 0.9.1-1421 (September 12, 2010)

• fixed a bug in -2 LL calculation in RAM models with definition variables and raw data

Release 0.9.0-1417 (September 10, 2010)

• improved error messages for non 1 x n means vectors in FIML and ML
• fixed a performance bug that was forcing too many recalculations of the covariance matrix in FIML optimizations.
• defualt behavior is to disable standard error calculations when model contains nonlinear constraints
• improved error messages for NA values in definition variables
• added error message when expected covariance dimnames and threshold dimnames do not contain the same elements.
• fixed a bug when mxRename() encounters a numeric or character literal.
• new chapters added to OpenMx user guide.
• fixed a bug in -2 log likelihood calculation with missing data

Release 0.5.2-1376 (August 29, 2010)

• improved error messages when identical label is applied to free and fixed parameters
• added ‘onlyFrontend’ optional argument to mxRun() function. See ?mxRun.
• disabling cbind() and rbind() transformations as they are broken.

Release 0.5.1-1366 (August 22, 2010)

• added error detection when multiple names are specified in mxMatrix(), mxAlgebra(), etc.
• removed ‘digits’ argument from mxCompare. Target behavior of argument was unclear. See ?mxCompare
• more informative error messages for ‘dimnames’ argument of objective functions
• more informative error messages when constraints have wrong dimensions
• improved error detected for ‘nrow’ and ‘ncol’ arguments of mxMatrix() function
• fixed a bug in ordinal FIML objective functions with non-used continuous data

Release 0.5.0-1353 (August 08, 2010)

• calculating cycle length of RAM objective functions
• bugfix: preserve rownames when converting data.frame columns to numeric values
• ‘nrow’ and ‘ncol’ arguments now supercede matrix dimensions in mxMatrix()
• add boolean argument ‘vector’ to mxRAMObjective() for returning the vector of likelihoods
• added demo(OneFactorModel_LikelihoodVector) as example of ‘vector=TRUE’ in RAM model
• cbind() and rbind() inside MxAlgebra expressions with all arguments as MxMatrix objects are themselves transformed into MxMatrix objects
• bugfix with square bracket substitution
• finishing implementing sorting of raw data in mxFIMLObjective()
• added ‘RAM Optimization’ and ‘RAM Max Depth’ to model options. See ?mxOption
• added support for linux x86_64 with gcc 4.1.x

Release 0.4.1-1320 (June 12, 2010)

• confidence interval optimizations now jitter if they can’t get started
• added error checking for dimensions of expected means in ML + FIML objectives
• check for missing observed means when using (optional) expected means in ML
• added citation(“OpenMx”) information

Release 0.4.0-1313 (June 09, 2010)

• fixed bug in calculation of confidence intervals around non-objective values
• checking for partial square bracket references on input
• fixed error reporting for non-positive-definite covariances in FIML
• implemented initial sorting-based speedup for FIML objectives
• added ‘No Sort Data’ to mxOptions()
• eliminated getOption(‘mxOptimizerOptions’) and getOption(‘mxCheckpointOptions’)
• added getOption(‘mxOptions’)
• error messages for illegal names provide function call information
• added ‘estimates’ column to confidence intervals in summary() of model
• fixed bug so checkpointing will work in R 2.9.x series
• fixed bug so mxRename() works on confidence interval specification
• renaming ‘estimates’ column of confidence interval summary output to ‘estimate’

Release 0.3.3-1264 (May 24, 2010)

• confidence interval frontend was requesting nonexistent matrices
• omxNewMatrixFromMxMatrix() assumed input was always integer vector S-expression
• confidence intervals mislabeling free parameter names

Release 0.3.2-1263 (May 22, 2010)

• added ‘newlabels’ argument to omxSetParameters() function
• now throwing errors to the user when detected from the backend in mxRun()
• checkpointing mechanism implemented - mxRun(model, checkpoint = TRUE)
• never computing confidence intervals for matrix cells where free = FALSE (all bets are off on algebras)
• added mxOption(model, “CI Max Iterations”, value)
• added documentation for mxRestore() function
• default expected means vectors are no longer generated

Release 0.3.1-1246 (May 09, 2010)

• new arguments to mxRun() for checkpointing and socket communication (doesn’t work yet)
• throw an error if FIML objective has thresholds but observed data is not a data.frame object.
• bugfix for R version 2.11.0 is detecting as.symbol(“”) character as missing function parameter
• mxFactor() function accepts data.frame objects
• added mxCI() function to calculate likelihood-based confidence intervals
• mxCompare() shows model information for base models
• added flag all=[TRUE|FALSE] to mxCompare() function
• ‘SaturatedLikelihood’ argument to summary() function will accept MxModel object

Release 0.3.0-1217 (Apr 20, 2010)

new features

• implemented new ordinal data interface http://openmx.psyc.virginia.edu/thread/416#comment-1421 (except for ‘means=0’ component)
• added mxFactor() function (see ?mxFactor for help)
• added R documentation for rvectorize() and cvectorize()
• implemented eigenvalues and eigenvectors
• added ‘numObs’, ‘numStats’ arguments to mxAlgebraObjective()
• added ‘numObs’, ‘numStats’ arguments to summary()

changes to interface

• mxConstraint(“A”, “=”, “B”) is now written as mxConstraint(A == B)
• renaming ‘cov’ argument to ‘covariance’ in omxMnor()
• renaming ‘lbounds’ argument to ‘lbound’ in omxMnor()
• renaming ‘ubounds’ argument to ‘ubound’ in omxMnor()
• renaming ‘cov’ argument to ‘covariance’ in omxAllInt()
• argument ‘silent = TRUE’ to mxRun() will no longer suppress warnings
• argument ‘suppressWarnings = TRUE’ to mxRun() will suppress warnings

bug fixes

• added error checking for mxBounds() with undefined parameter names
• improving error messages in mxMatrix() function
• fixed aliasing bug in mxAlgebra() with external variables and constant values

internal

• added directory to repository for nightly tests (“make nightly”)
• performance improvement to namespace conversion
• changing MxPath data structure from a list to an S4 object
• new signature for omxSetParameters(model, labels, free, values, lbound, ubound, indep)
• checked in implementation of mergesort
• changed mxCompare() signature to mxCompare(base, comparison, digits = 3)

Release 0.2.10-1172 (Mar 14, 2010)

• bugfix for assigning default data name when default data does not exist
• bugfix for sharing/unsharing data to a three-level hierarchy
• bugfix for error reporting in bad matrix access, uncalculated std. errors, and poor omxAllint thresholds
• implemented rvectorize, cvectorize algebra functions: vectorize by row, and vectorize by column
• not allowing the following forbidden characters in names or labels: “+-!~?:*/^%<>=&|$”
• added timestamp to summary() output
• reimplemented summary() function to handle unused data rows, and independent submodels
• reimplemented names(model), model$foo, and model$foo <- value to return all components of a model tree
• started work on an “OpenMx style guide” section to the User Guide.
• fix documentation + demo errors brought to our attention by dbishop

Release 0.2.9-1147 (Mar 04, 2010)

• bugfix for ordinal FIML with columns that are not threshold columns
• bugfix for detection of algebraic cycles when multiple objective functions are present
• test cases included for standard error calculation
• enabled standard error calculation by default

Release 0.2.8-1133 (Mar 02, 2010)

• bugfix for memory leak behavior in kronecker product calculation
• implemented kronecker exponentiation operator %^%.
• actually updating means calculations in ordinal FIML models
• removing standard errors from summary() until they are computed correctly

Release 0.2.7-1125 (Feb 28, 2010)

• added ‘unsafe’ argument to mxRun function. See ?mxRun for more information.
• bugfix for filtering definition variable assignment to current data source.
• bugfix for using data.frame with integer type columns that are not factors.

Release 0.2.6-1114 (Feb 23, 2010)

• implemented omxAllInt, use ?omxAllInt for R help on this function
• implemented omxMnor, use ?omxMnor for R help on this function
• added option to calculate Hessian after optimization. See ?mxOption for R help.
• added option to calculate standard errors after optimization. See ?mxOption for R help.
• added R documentation for vech, vechs, vec2diag, and diag2vec
• performance improvements for independent submodels
• added ‘silent=FALSE’ argument to mxRun() function
• added R documentation for omxApply(), omxSapply(), and omxLapply()
• wrote mxRename() and added R documentation for function
• added ‘indep’ argument to summary() to ignore independent submodels (see ?summary)
• added ‘independentTime’ to summary() output. Wall clock time for independent submodels.
• added ‘wallTime’ and ‘cpuTime’ to summary() output. Total wall clock time and total cpu time.
• implemented ‘:’ operator for MxAlgebra expressions. 1:5 returns the vector [1,2,3,4,5]
• implemented subranges for ‘[‘ operator in MxAlgebra expressions. foo[1:5,] is valid inside algebra.
• added omxGetParameters, omxSetParameters, omxAssignFirstParameters. Use ? for documentation.
• renamed all objective function generic functions from omxObj* to genericObj*
• enumerated OpenMx and NPSOL options in ?mxOption documentation
• implemented foo[x,y] and foo[x,y] <- z for MxMatrix objects

Release 0.2.5-1050 (Jan 22, 2010)

• added ‘mxVersion’ slot to output of summary() function.
• added documentation of summary() function.
• set default function precision for ordinal FIML evaluation to “1e-9”
• not throwing an error on mxMatrix(‘Full’, 3, 3, labels = c(NA, NA, NA))
• applying identical error checking to single thresholds matrix and single thresholds algebra
• implemented generic method names() for MxModel objects.
• implemented vec2diag() and diag2vec() matrix algebra functions.

Release 0.2.4-1038 (Jan 15, 2010)

• definition variables can now be used inside algebra expressions
• definition variables inside of MxMatrices will populate to the 1st row before conformability checking. In plain english: you do not need to specify the starting values for definition variables.
• the square-bracket operator when used in MxMatrix labels is no longer restricted to constants for the row and column. The row and column arguments will accept any term that evaluates to a scalar value or a (1 x 1) matrix.
• summary() on a model returns a S3 object. Behaves like summary() in stats package.
• eliminated UnusualLabels.R test case. Too many problems with windows versus OS X versus linux.
• implemented vech() and vechs() functions: half-vectorization and strict half-vectorization
• fixed bug in ordinal FIML when # of data columns > # of thresholds
• added ‘frontendTime’ and ‘backendTime’ values to summary() output. They store the elapsed time of a model in the R front-end and C back-end, respectively.
• created a name space for the OpenMx library. Only mx**() and omx**() functions should be exported to the user, plus several miscellaneous matrix functions and S4 generic functions.
• corrected ‘observedStatistics’ output of summary() to exclude definition variables
• corrected ‘observedStatistics’ output of summary() count the number of equality constraints

Release 0.2.3-1006 (Dec 04, 2009)

• added ‘vector’ argument to mxFIMLObjective() function. Specifies whether to return the likelihood vector (if TRUE) or the sum of log likelihoods (if FALSE). Default value is FALSE.
• renamed omxCheckEquals() to omxCheckIdentical(). omxCheckIdentical() call “identical” so that NAs can be compared.
• added checking of column names of F and M matrices in RAM objective functions.
• added ‘dimnames’ argument to mxRAMObjective() function. Populates the column names of F and M matrices.
• added square bracket operator to MxAlgebra expressions. A[x,y] or A[,y] or A[x,] or A[,] are valid.
• square bracket operator supports row and column string arguments.
• mxModel(remove=TRUE) accepts both character names or S4 named entities.
• added support for x86_64 on OS X 10.6 (snow leopard)
• fixed support for x86_64 on Ubuntu 9.10 (gcc 4.4)
• throw error message when inserting a named entity into a model with an identical name
• added Anthony William Fairbank Edwards “Likelihood” (1972; 1984) A, B, O blood group example to online documentation

Release 0.2.2-951 (Oct 29, 2009)

• omxGraphviz() either prints to stdout or to a filename
• updated omxGraphviz() to draw an arrow if (value != 0 || free == TRUE || !is.na(label))
• omxGraphviz() returns a character string invisibly
• error checking for bogus definition variables
• summary() uses matrix dimnames by default, use options(‘mxShowDimnames’=FALSE) to disable
• added support for gcc 4.4 (Ubuntu 9.10) on x86 and x86_64 architectures
• created R documention for omxGraphviz() function
• generalized dependency specification for objective functions (omxObjDependencies)
• fixed cross-reference links in User Guide

Release 0.2.1-922 (Oct 10, 2009)

• checked observed data for dimnames on ML objective functions
• using dimnames= argument to mxMLObjective() and mxFIMLObjective() propagates to MxMatrix objects on output.
• bug fix for error message where model name is incorrect
• updated user guide in response to feedback from beta testers
• incremented version number to 0.2.1-922 to sync demos and user guide

Release 0.2.0-905 (Oct 06, 2009)

• several of the twin model demo examples have been recoded.
• fixed bug with non-floating point matrices.
• more error checking for mxPath().
• tools/mxAlgebraParser.py will convert Mx 1.0 algebra expressions (Python PLY library is required).
• renamed “parameter estimate” column to “Estimate” and “error estimate” to “Std.Error” in summary().
• added ‘dimnames’ argument to mxFIMLObjective() and mxMLObjective().
• error checking for RAM models with non-RAM objective functions.

Release 0.1.5-851 (Sep 25, 2009)

• improved error messages on unknown identifier in a model (beta tester issue)
• fixed bug in mxMatrix() when values argument is matrix and byrow=TRUE
• implemented square-bracket substitution for MxMatrix labels
• fixed a bug in computation of omxFIMLObjective within an algebra when definition variables are used
• significant alterations to back-end debugging flags
• tweaked memory handling in back-end matrix copying
• added support for x86_64 linux with gcc 4.2 and 4.3

Release 0.1.4-827 (Sep 18, 2009)

• added checking and type coercion to arguments of mxPath() function (a beta tester alerted us to this)
• moved matrices into submodels in UnivariateTwinAnalysis_MatrixRaw demo
• added Beginners Guide to online documentation
• mxRun() issues an error when the back-end reports a negative status code
• named entities and free or fixed parameter names cannot be numeric values
• constant literals are allowed inside mxAlgebra() statements, e.g. mxAlgebra(1 + 2 + 3)
• constant literals can be of the form 1.234E+56 or 1.234e+56.
• type checking added to mxMatrix arguments (prompted by a forum post)
• mxPath() issues an error if any of the arguments are longer than the number of paths to be generated
• data frames are now accepted at the back-end
• FIML ordinal objective function is now working. Still a bit slow and inelegant, but working
• FIML ordinal now accepts algebras and matrices. dimnames of columns must match data elements
• implemented free parameter and fixed parameter substitution in mxAlgebra statements
• implemented global variable substitution in mxAlgebra statements
• turned off matrix and algebra substitution until a new proposal is decided
• snow and snowfall are no longer required packages
• added cycle detection to algebra expressions
• mxEval() with compute = TRUE will assign dimnames to algebras
• added dimnames checking of algebras in the front-end before optimization is called
• added ‘make rproftest’ target to makefile

Release 0.1.3-776 (Aug 28, 2009)

• mxEvaluate() was renamed to mxEval() after input from beta testers on the forums.
• new function mxVersion that prints out the current version number (beta tester request).
• When printing OpenMx objects, the @ sign is used where it is needed if you would want to print part of the object (beta tester request).
• now supports PPC macs.
• implemented AIC, BIC and RMSEA calculations.
• mxMatrix documentation now talks about lower triangular matrices (beta tester request).
• fixed bugs in a number of demo scripts.
• added chi-square and p-value patch from beta tester Michael Scharkow.
• added comments to demo scripts.
• fixed a bug in the quadratic operator (a beta tester alerted us to this).
• means vectors are now always 1xn matrices (beta tester request).
• added an option “compute” to mxEval() to precompute matrix expressions without going to the optimizer.
• Matrix algebra conformability is now tested in R at the beginning of each mxRun().
• named entities (i.e. mxMatrices, mxAlgebras, etc.) can no longer have the same name as the label of a free parameter. (This seems obscure, but you will like what we do with it in the next version!)
• can use options(mxByrow=TRUE) in the R global options if you always read your matrices in with the byrow=TRUE argument. Saves some typing. (beta tester request)
• fixed the standard error estimates summary.
• added mxVersion() function to return the version number (as a string).

Release 0.1.2-708 (Aug 14, 2009)

• Added R help documentation for omxCheckCloseEnough(), omxCheckWithinPercentError(),

  omxCheckTrue(), omxCheckEquals(), and omxCheckSetEquals()

• (mxMatrix) Fixed a bug in construction of symmetric matrixes.

  • now supports lower, standardized, and subdiagonal matrices.

Release 0.1 (Aug 03, 2009)

• (mxEvaluate) mxEvaluate translates MxMatrix references, MxAlgebra references, MxObjectiveFunction references, and label references.

• (mxOptions) added ‘reset’ argument to mxOptions()

• (mxPath) renamed ‘start’ argument of mxPath() to ‘values’

  • renamed ‘name’ argument of mxPath() to ‘labels’
  • renamed ‘boundMin’ argument of mxPath() to ‘lbound’
  • renamed ‘boundMax’ argument of mxPath() to ‘ubound’
  • eliminated ‘ciLower’ argument of mxPath()
  • eliminated ‘ciUpper’ argument of mxPath()
  • eliminated ‘description’ argument of mxPath()

• (dimnames) implemented dimnames(x) for MxMatrix objects

  • implemented dimnames(x) <- value for MxMatrix objects
  • implemented dimnames(x) for MxAlgebra objects
  • implemented dimnames(x) <- value for MxAlgebra objects

• (mxMatrix) added ‘dimnames’ argument to mxMatrix()

• (mxData) renamed ‘vector’ argument of mxData() to ‘means’