search

PrincetonUniversity / princeton-mvpa-toolbox

Automatically exported from code.google.com/p/princeton-mvpa-toolbox
GNU General Public License v3.0
64 stars 56 forks source link

MVPA toolbox did not account for the probability of underflow problem #5

Open GoogleCodeExporter opened 8 years ago

GoogleCodeExporter commented 8 years ago 
What steps will reproduce the problem?
1. subj = load_afni_mask(subj,'wholebrain','wholebrain+orig');

2. class_args.train_funct_name = 'train_gnb';
   class_args.test_funct_name = 'test_gnb';
   class_args.nHidden = 0;
3.

What is the expected output? What do you see instead?
The output I see is: Perf = [ 0.4793 0.4793 0.4793 0.4793     0.4793     0.4793 
0.4793 0.4793 0.4793 0.4793]
Perf_total = 0.4793 and I noticed the value of "acts" is all
NaN, please see the attachment!

What version of the product are you using? On what operating system?
Matlab R2014a.  Windows 7 (64bit)

Please provide any additional information below.
I just simply modify the "tutorial_easy.m" as follows
1) I change the mask to the intra-cranial"wholebrain" mask.
2) I change the Backprop Classifier to Gaussian Naive Bayes Classifier.
Then I run the code and find that MVPA toolbox did not account for the 
possibility of underflow.
I paste part of the result here. Please take a look at it!

Original issue reported on code.google.com by smithli2...@gmail.com on 29 Jan 2015 at 10:13

Attachments: