extracting "importance" maps

[Jong Yoon]

I would like to extract "importance" maps as it was conducted in the
Polyn '05 study. Does anyone have more detailed documentation of
exactly how to construct these maps?

Thanks in advance.

jong

[Greg Detre]

dear jong,

that's a good question. i had a look around, and i don't
*think* we'd ever put the script to do that online. i've
added it now, so it's available in the SVN development
version, or you can just download it directly here

https://www.csbmb.princeton.edu/repos/distpat/trunk/mvpa/core/learn/interpret_weights.m

it's relatively complicated, since it tries to do as much of
the work for the user as possible, using the RESULTS
structure to guess which SUBJ structure objects to pay
attention to

hopefully the documentation will tell you all you need to
know. as should be clear, it only works for backprop with no
hidden layer.

you may find that you just want to grab the key lines that
grab the weights from the network structure, e.g.

curWts=results{i}.iterations(j).scratchpad.net.IW{1}(k,:)';

and the ones below that do the main computation

i should say that i haven't looked at this script in a long
time. we do have a unit test for it though, and i'll try to
put that online too sometime soon, to help you test and play
with it

g

--

---
Greg Detre
cell: 617 642 3902
email: gr...@gregdetre.co.uk
web: http://www.princeton.edu/~gdetre/

[Greg Detre]

dear jong,

on closer inspection, i don't think our unit test will help
you very much.

but let us know how you get on with the importance maps,

g

[Greg Detre]

One final point to note - we've created a wiki page on this
subject. If things get updated, the changes will show up there.

https://compmem.princeton.edu/mvpa_docs/TutorialImportanceMaps

g

[qasim....@gmail.com]

Hello everybody

I have a quick question

Can I use interprets_weights.m to calculate the importance maps of my Random Forest classifier ??? I read here that people have used it with back propagation; but I was wondering it could also be used with random forest etc. !! 

[subj] = interpret_weights(subj,results,varargin)

what are my results here ?? Results of classification from my random forest ? But I really wonder how does it consider then which of the particular weights were weighted high by the classifier because we doesnt seem to input the learning from random forest, do you understand what I mean ?

best,

Qasim