fitlme "Index exceeds matrix dimensions."

Hi, I'm trying to use the new fitlme bult-in function (R2015a) and it keeps going on an error:

Index exceeds matrix dimensions.

Error in classreg.regr.LinearLikeMixedModel.makeSparseZ (line 1312)
                idxr     = accumarray(subs,vals(I),[],@(x) {x});

Error in LinearMixedModel/fitStandardLMEModel (line 1225)
        Zs = LinearMixedModel.makeSparseZ(Z,q,lev,Gid,N);

Error in LinearMixedModel/fitter (line 822)
        model.slme = fitStandardLMEModel(model);

Error in classreg.regr.FitObject/doFit (line 220)
            model = fitter(model);

Error in LinearMixedModel.fit (line 2395)
        model = doFit(model);

Error in fitlme (line 224)
    lme = LinearMixedModel.fit(ds,formula,varargin{:});

Also, running the fnuction takes too long (the error appears after several minutes of wating) and consume a lot of RAM. This is the line I'm running:

lmm1 = fitlme(datos_tabla, 'E1_T1 ~ pred + (1|SUJ)' );

Answer 1

[Gautam Pendse]

Hi Bruno,

This doesn't look like the intended behavior but it is difficult to say why you are getting that error. Could you post your data that reproduces the problem?

Gautam

[Bruno Bianchi]

Thanks for the answer, here I update the table that reproduce that message.

The table I'm uploading is a part of the complete table I have, the name of the original es datos_tabla, the name of this table es tabla_export, so the line that reproduce the message is:

lmm1 = fitlme(tabla_export, 'E1_T1 ~ pred + (1|SUJ)' );

[Gautam Pendse]

Hi Bruno,

Your table is of size 1-by-3 with 3 predictors. Each variable in the table is of size 1-by-N:

The error message can be improved but fitlme expects your table to be of size N-by-3 where each row represents 1 observation. In the code below, I modify your table by transposing variables and then fit the model:

%%Display the original table
test         = load('tabla_export.mat');
tabla_export = test.tabla_export;

size(tabla_export)

tabla_export

% Variables in your table are of size 1-by-N. The expectation is that your
% table will have N rows and each row would represent one observation. You
% can do this by just transposing the variables in your table as shown
% below.

%%Transpose 1-by-N variables into N-by-1 variables
E1_T1 = tabla_export.E1_T1;
pred  = tabla_export.pred;
SUJ   = tabla_export.SUJ;

E1_T1        = E1_T1';
pred         = pred';
SUJ          = SUJ';
newtbl       = table();
newtbl.E1_T1 = E1_T1;
newtbl.pred  = pred;
newtbl.SUJ   = SUJ;

%%Fit the model
rng(0,'twister')
lme = fitlme(newtbl,'E1_T1 ~ pred + (1|SUJ)','verbose',1,'startmethod','random')

[Bruno Bianchi]

That really works! Thank you very much, and I hope the problem to be solve in the future!