Check dimension conformability of all arguments

[mhunter1]

Original report by Michael Hunter (Bitbucket: mhunter, GitHub: mhunter).

---

A user created a seg fault by having 1 observed variable but a 4x4 measurement noise covariance.

[mhunter1]

Here's the conformability check function from OpenMx

checkSSMConformable <- function(mat, rows, cols, matname, modname){
	if( nrow(mat) != rows || ncol(mat) != cols ){
		msg <- paste("The ", matname, " matrix is not the correct size",
			" in the state space expectation of model ", modname,
			".  It is ", nrow(mat), " by ", ncol(mat), " and should be ",
			rows, " by ", cols, ".", sep="")
		stop(msg, call. = FALSE)
	}
}

[mhunter1]

Here's how the matrices get checked.

		# Conformability checking
		ldim <- ncol(Cmatrix)
		mdim <- nrow(Cmatrix)
		checkSSMConformable(Amatrix, ldim, ldim, 'A', omxQuotes(modelname))
		checkSSMConformable(Cmatrix, mdim, ldim, 'C', omxQuotes(modelname))
		checkSSMConformable(Qmatrix, ldim, ldim, 'Q', omxQuotes(modelname))
		checkSSMConformable(Rmatrix, mdim, mdim, 'R', omxQuotes(modelname))
		checkSSMConformable(Xmatrix, ldim, 1, 'x0', omxQuotes(modelname))
		checkSSMConformable(Pmatrix, ldim, ldim, 'P0', omxQuotes(modelname))

[symiin]

In dynr.model, check conformability of the followings:

1. dynamics (Meng and Hui-Ju)
   -Check the length of the list of formulas. Compare length to the length of list in prep.initials.

• Within each set of formulas, compare number of formulas against length of state.names from measurement model.
  -Check length of startvals to see if it matches the length of parameters after name checking

2. noise (Yanling and Jonathan)

• Check dimension of measurement noise covariance matrix to see if it matches length of obs.names
  -Check dimension of process noise covariance matrix against number of formulas and also length of state.names from prep.measurement.

3. initial (Mike and Jungmin)
   -Check number of rows in initial condition mean vector to see if it matches the length of state.names
   -Check dimension of initial condition covariance matrix to see if the dimension matches the length of state.names
   -In initial: Check the length of the initial regime probabilities vector to see if it matches the length of the list of elements in measurement; in dynamics

4. regimes (Linying and Sy-Miin)

• Check the dimension of the regime transition logodds matrix to see if it is the right dimension with and without covariates
• Check the dimension against the lengths of lists of elements in measurement and dynamics

[mhunter1]

Maybe this is a trivial distinction, but I recommend a slightly different strategy. Instead of checking everything against everything else, use the strategy outlined below.

Here's how the matrices get checked.

		# Conformability checking
		ldim <- ncol(Cmatrix)
		mdim <- nrow(Cmatrix)
		checkSSMConformable(Amatrix, ldim, ldim, 'A', omxQuotes(modelname))
		checkSSMConformable(Cmatrix, mdim, ldim, 'C', omxQuotes(modelname))
		checkSSMConformable(Qmatrix, ldim, ldim, 'Q', omxQuotes(modelname))
		checkSSMConformable(Rmatrix, mdim, mdim, 'R', omxQuotes(modelname))
		checkSSMConformable(Xmatrix, ldim, 1, 'x0', omxQuotes(modelname))
		checkSSMConformable(Pmatrix, ldim, ldim, 'P0', omxQuotes(modelname))

That is, set some ground truth for

1. the number of regimes
2. the number of latent variables
3. the number of observed variables
4. the names of the latent variables
5. the names of the observed variables

Then check that everything that should match this ground truth does in fact match it.

As evidenced by the code I wrote for OpenMx, I like the idea of having the measurement model (e.g. prep.measurement) define the number and names for both latent and observed variables.

Set ground truths

• Set number of regimes nr
• Set number of latent variables ne
• Set names of latent variables nne
• Set number of observed variables ny
• Set names of observed variables nny
• Set the names and numbers of covariates

Dynamics (@meng-chen-2013 @hjhung)

• Check that prep.formulaDynamics formulas have ne left-hand sides
• Check that prep.formulaDynamics formulas use all and only nne in left-hand sides
• Check that prep.matrixDynamics is ne by ne

Noise (@yanlingli1 @JPark93)

• Check that measurement noise is ny by ny
• Check that dynamic noise is ne by ne

Initial (@mhunter1 @ares7823)

• Check that ini.state is ne by 1
• Check that ini.cov is ne by ne
• Check that regimep is nr by 1

Regimes (@symiin linying)

• Check that the dimension of the regime transition logodds matrix is the right dimension with and without covariates
• Check that nr does not contradict the number of regimes in dynamics, noise, or initial

[mchdfs]

I put in an additional check for the class of dynamic argument in dynr.model:

if(!class(dynamics) %in% c("dynrDynamicsFormula","dynrDynamicsMatrix")){
stop("Check to see that dynamics argument is of the correct class. Hint: it should be either 'dynrDynamicsFormula' or 'dynrDynamicsMatrix'.")
}

in case people put in a set of formulae instead.

[mhunter1]

@meng-chen-2013 This is a nice idea. I think we should go all the way with it. See #248.