Username Remember Me?
Password   forgot password?
   
   
Deep learning question
Posted: 23 February 2017 01:42 PM   [ Ignore ]  
Master
RankRankRank
Total Posts:  55
Joined  2010-05-27

Dear Developer,

I was testing the sample codes in the manual and encountered the following error. can you please make your suggestions?

%%%%%
load nist16.mat
a.imsize
a=setprop(a,’imsize’,[16 16],’data’)
[tr,ts]=randsubset(a,0.5)
p=sddeepnet(tr,’conv’,[5 5 1 10],’bnorm’,’mpool’,3,’relu’,’conv’,[5 5 10 20],’bnorm’,’mpool’,2,’relu’,’conv’,[5 5 20 10],’epoch’,20)
%%%%%%%%%%%

The error message:

>> p=sddeepnet(tr,’conv’,[5 5 1 10],’bnorm’,’mpool’,3,’relu’,’conv’,[5 5 10 20],’bnorm’,’mpool’,2,’relu’,’conv’,[5 5 20 10],’epoch’,20)
ind acion input -> output :  filter count step
----------------------------------------------------------------------------------
1. ‘conv’ :  16x16x1 12x12x10 5x5x1 10 1
2. ‘bnorm’ :  12x12x10 12x12x10
3. ‘mpool’ :  12x12x10 10x10x10 3x3 1
4. ‘relu’ :  10x10x10 10x10x10
5. ‘conv’ :  10x10x10 6 x6 x20 5x5x10 20 1
6. ‘bnorm’ :  6 x6 x20 6 x6 x20
7. ‘mpool’ :  6 x6 x20 5 x5 x20 2x2 1
8. ‘relu’ :  5 x5 x20 5 x5 x20
9. ‘conv’ :  5 x5 x20 1 x1 x10 5x5x20 10 1
Architecture OK
Attempt to execute SCRIPT vl_nnconv as a function:
C:\APR\perClass_Pro_5.0\perclass\matconvnet\matlab\vl_nnconv.m

Error in vl_simplenn (line 296)
res(i+1).x = vl_nnconv(res(i).x, l.weights{1}, l.weights{2}, ...

Error in dn_process

Error in dn_process

Error in sddeepnet

thank you!

Profile
 
 
Posted: 23 February 2017 04:48 PM   [ Ignore ]   [ # 1 ]  
Administrator
Avatar
RankRankRankRank
Total Posts:  360
Joined  2008-04-26

Dear Liwei,

this seems like path problem. It might happen, if you put perclass or matconvnet on Matlab path “with all sub-folders”. The only directory you really need to put manually on Matlab path is perclass (not matconvnet - it is bundled with perClass and should be put on the path automatically).

Second situation where we can see such errors is if people change their working directory to the matconvnet folder

Does it help?

With Kind Regards,

Pavel

Profile
 
 
Posted: 27 February 2017 10:06 PM   [ Ignore ]   [ # 2 ]  
Master
RankRankRank
Total Posts:  55
Joined  2010-05-27

thanks! it works.

However, the document says: it is possible to use a custom- built Matconvnet. In order to do so, use “noaddpath” when invoking sddeepnet and add the following directories of Matconvnet to Matlab path:

matconvnet/matlab;
matconvnet/matlab/compatibility/parallel
matconvnet/matlab/mex
matconvnet/matlab/simlenn

are these folders under teh PerClass folder?

thank you!

Image Attachments
CustomMatconvnetInstallation.PNG
Profile
 
 
Posted: 28 February 2017 10:26 AM   [ Ignore ]   [ # 3 ]  
Administrator
Avatar
RankRankRankRank
Total Posts:  360
Joined  2008-04-26

if you have Parallel computing toolbox, do not put the matconvnet/matlab/compatibility/parallel directory on your Matlab path.

This directory is provided by matconvnet toolbox which perClass uses for training deep convolutional networks. It is provided for situations, where parallel computing toolbox is not present and deep net training will be handled by CPU only. The labindex and numlabs functions from parallel toolbox are then replaced by dummy versions from matconvnet.

I’d recommend to first remove all perclass and matconvet paths from Matlab path, add perclass and then the paths above only without matconvnet/matlab/compatibility/parallel

Does it help?

With Kind Regards,

Pavel

Profile
 
 
Posted: 28 February 2017 04:46 PM   [ Ignore ]   [ # 4 ]  
Master
RankRankRank
Total Posts:  55
Joined  2010-05-27

thanks! this way works!

Profile
 
 
   
 
 
‹‹ Deep learning test      Threshold SVC ››