micheletufano / autoencode Goto Github PK

Hi
I had trained a corpus using your tool. Now I want to test another corpus using the same model. So for that, I made a slight modification to your code in ComputeSimilarities.m.

Instead of :
load(strcat(ODIR,'/data.mat'));
I changed it to :
[corpora,labels,vocabulary,We] = preprocess(IDIR,ODIR,MAX_SENTENCE_LENGTH);
so that it calculates data.mat and gives me the attributes. After I run the modified ComputeSimilarities.m, I get the following error:

Please let me know if I am missing out something.
Thanks.

When I run run_rae.sh on my custom dataset, it appears:

Starting parallel pool (parpool) using the 'local' profile ...
Connected to the parallel pool (number of workers: 6).
 Iteration   FunEvals     Step Length    Function Val        Opt Cond
Step direction is illegal!

Position train (line 19)
[opttheta,cost] = minFunc(@(p)costfunc(p, ...

Position main (line 36)
[opttheta,cost,freq] = train(ODIR, ...

More specifically, I observe that the variable d is an array of NaN (line 938, minFunc.m), which is reported to be illegal.

Why this situation happens? And how to deal with this problem?