MLP classifier resulting in 100% positive ROI results in ROITagger

[GoogleCodeExporter]

What steps will reproduce the problem?
1. Invoke MLP classifier in ROI tagger logic
2. When model goes to be trained on ROIs, 
3. Classifies all results as positive.

What is the expected output? What do you see instead?
on any given ROI set, result should match the known training set.

expected, seven "1" and seven "0"
out of 14 ROIs, 

instead all "1"

Original issue reported on code.google.com by Jon.Sand...@gmail.com on 21 Nov 2013 at 10:03

Attachments:

• MLP_incorrect_log.txt

[GoogleCodeExporter]

This is related to annotool.analysis.ThreeDROIAnnotation.
It is a separate way of training and testing 3DROIs for the 3D object counting 
project.  Inside the class, it builds an arraylist of 3D ROIs for 
training/testing and also does postprocessing such as meanshifting. It can be 
invoked from a menu called ROITagger. When MLPClassifier is used in that 
context,  it breaks.

Original comment by jzho...@gmail.com on 22 Nov 2013 at 7:12