GIS vs QNTrainer

["Russ, Daniel (NIH/CIT) [E]" <dr...@mail.nih.gov>]

Hello,
   I was looking to switch from the GIS trainer to the QN trainer.  From what I understand, both methods just attempt to solve the maxent weight optimization problem.  I was expecting the results to be similar.’

Here is my made up training data:

OUTCOME features
A feature1
B feature1 feature2

Testing data:
? feature1
? feature2
? feature1 feature2

results:

feature1
GIS: A: 0.951 B:0.049
QN: A: 0.610 B:0.390

feature2
GIS: A: 0.002 B:0.998
QN: A: 0.169 B:0.830

feature1 feature2
GIS: A: 0.039 B:0.961
QN: A: 0.242 B:0.758

Is this the expected result?  Should the two training model give such divergent results?  I understand the dataset is trivial, but I think these differences are huge.
Any advice?

Thank you,
Daniel

Daniel Russ, Ph.D.
Staff Scientist, Office of Intramural Research
Center for Information Technology
National Institutes of Health
U.S. Department of Health and Human Services
12 South Drive
Bethesda,  MD 20892-5624