visionlab / memtoolbox Goto Github PK

When attempting to run the MemToolbox demo experiment "ColorWorkingMemoryExperiment" I encounter the following error:

**Cannot find an exact (case-sensitive) match for 'RandSample'

Error in ColorWorkingMemoryExperiment (line 40)
itemToTest(trialIndex) = RandSample(1:nItems);**

Is there a "RandSample.m" file required to run this function that is different from the one provided by MathWorks ?

Many thanks,

Luc Boutsen
Aston University, Birmingham, UK

We might consider try/catch-ing on the line that calls nansum() in sumWithNans(), to note the most likely source of the error is a missing Statistics Toolbox.

perhaps add it as an optional parameter to MemFit

"The issue of whether or not to include a Bayes factor computation using a sampling method is, frankly, a bit of a conundrum. I am not entirely comfortable with it, as my own experience with calculating Bayes' factors is mixed. Even a Laplace approximation, as suggested by the reviewer as an alternative, can be highly unstable for certain models. I would urge you not to include it as a pre-packaged model comparison tool, but I will not insist on it. You should, however, include a brief, critical discussion of the problems associated with computing Bayes' factors in the paper. "

for example,

AIC = ModelComparison_AIC_BIC(MemDataset(3), {SwapModel(), StandardMixtureModel()})

would report:

Error using mle>validateInputTypes (line 769)
Expected input to be one of these types:

single, double

Instead its type was struct.

Error in mle (line 258)
validateInputTypes(data, cens, freq, alpha, ntrials, theta, mu, truncation, start);

Error in MLE (line 42)
vals{c} = mle(data, 'logpdf', model.logpdf, 'start', model.start(c,:), ...

Error in ModelComparison_AIC_BIC (line 69)
[params, logLike(md)] = MLE(data, models{md});

Now that we have release a second version of the MemToolbox, we should update the documentation to recommend that citations include the version number used.

Hi, I am trying to use the memtoolbox to model my data, but I run into a problem everytime I use the function MemFit (even when I use the example data). Namely, I get this error:

`Just a moment while MTB fits a model to your data...

Running 3 chains...
Error using mvnrnd (line 35)
Index exceeds the number of array elements (0).

Error in MCMC>MCMC_Chain (line 234)
movement = mvnrnd(zeros(1, length(startInfo.cur)), curCov);

Error in MCMC (line 141)
parfor c=1:numChains

Error in MemFit>MemFit_SingleData (line 188)
posteriorSamples = MCMC(data, model, 'Verbosity', verbosity-1, ...

Error in MemFit (line 117)
fit = MemFit_SingleData(data, model, verbosity);`

I tried many different solutions, but could not find a way to solve this. I am using Matlab 2019a, could it be that some functions have changed since the first publication of the toolbox?

Many thanks! :)

The repmat function in the lightspeed toolbox is incompatible with MemToolbox because it only handles matrices, not cell arrays, as is required by functions such as Hierarchical, which use repmat to create cell arrays of parameter names.

Hi,

In my experiment I have 3 conditions, is there a way to plot G and precision for all three of them in one plot?

Thank you!
Best
Ilaria

The Parallel Computing Toolbox fails with the latest Java update:
http://www.mathworks.com/support/bugreports/919688

It produces this error:

Error using matlabpool (line 134)
Java exception occurred:
java.lang.NullPointerException
    at java.util.logging.Logger.demandLogger(Logger.java:286)
    at java.util.logging.Logger.getLogger(Logger.java:321)

Error in aspenfun (line 3)
if(~(matlabpool('size') > 0)) matlabpool open;

It would be nice to catch the error and inform MemToolbox users so they don't despair.

One way to do it is by checking:

version -java

"Minor Point: Page 5. The authors claim that Jeffrey's priors "limit the range of parameter to those that are meaningful." Quite the opposite. Jeffrey's priors are location and/or scale invariance, that is, they do NOT limit the range of parameters. "

Hi! I wanted to use the Hierarchical function to create a hierarchical model for my project, but got an error message:

Error using mle>validateInputTypes (line 769)
Expected input to be one of these types:

single, double

Instead its type was struct.

After digging into it, it seems to me that in MAP(data, model) function, which is used in Hierarchical , the mle function from Statistics and Machine Learning Toolbox is used, with:

vals{c} = mle(data, 'logpdf', logPosterior, 'start', model.start(c,:), ...
'lowerbound', model.lowerbound, 'upperbound', model.upperbound, ...
'options', options);

But the mle is expecting "data" to be a vector variable, not a struct.

I'm not sure if it is a compatibility problem caused by the newer version of Statistics Toolbox (mine is version 12.2). Could you please see into this and update the MAP() function accordingly? It would be great if there's another way to run the MAP() function because I desperately need it for my group-level analysis - I tried fitting each subject individually but the results are not great.

Thank you!

Best,
Jiangang

"You should improve the discussion of the different model comparison metrics (see reviewer 2's comments). I agree with the reviewer that simply choosing the method least advantageous to one's preferred model is unprincipled. I don't expect a lengthy analysis of AIC, BIC, DIC and AICc, but a bit more than is currently provided is needed to guide users. The current discussion is a bit shallow as it distinguishes the different methods primarily based on their apparent biases rather than on the different computational principles from which they are derived."

"C. Minor Point: One of the earliest hierarchical models of psychological process is Rouder et al. (2003, Psychometrika). Rouder and Lu (2005, PBR) provide an early tutorial review of the benefits of hierarchical models of psychological process. The citations would be appropriate."

If you have set size 2, and try to run a swap model, it fails to converge.

demo10.m is currently:

data.errors = [-12 3 38 27 -29 21 -22 52 -2 -19 21 17 38 6 34 25 44];
MemFit(data, Orientation(WithBias(StandardMixtureModel)))

which is missing one of Orientation's required parameters, whichParameters, resulting in this error:

>> MemFit(data, Orientation(WithBias(StandardMixtureModel)))
??? Input argument "whichParameters" is undefined.

Error in ==> Orientation at 33
  model.upperbound(whichParameters) = model.upperbound(whichParameters) ./ 2;

But when I do specify whichParameters, the MCMC chains never converge:

>> data.errors = [-12 3 38 27 -29 21 -22 52 -2 -19 21 17 38 6 34 25 44];
MemFit(data, Orientation(WithBias(StandardMixtureModel), 1))

Mean percent correct: 0.75
          Model:   Standard mixture model with bias (for orientation)
     Parameters:   mu, g, sd

Just a moment while MTB fits a model to your data...

   Running 3 chains...
   ... not yet converged (200); btw/within variance: Inf
   ... not yet converged (400); btw/within variance: Inf
   ... not yet converged (600); btw/within variance: Inf

Hey guys,

Thanks for the toolbox. Great stuff. I noticed a little typo in the help of the VariablePrecisionModel function. When you talk about the optional argument 'HigherOrderDist', the examples pass this argument to the StandardMixtureModel and not the VariablePrecisionModel, which you probably mean.

Hope this helps, thanks again for the great toolbox!

Wouter