Hi all,<br><br>I'm still trying to get recursive feature selection to work with the example from <a href="http://dev.pymvpa.org/featsel.html#recursive-feature-elimination">http://dev.pymvpa.org/featsel.html#recursive-feature-elimination</a>, however I can't get past an error from the line with <code class="python spaces"></code><code class="python plain">confusion_state</code><code class="python keyword">=</code><code class="python string">"confusion"</code><code class="python plain">.<br>
<br></code> Here's the code: <br>rfesvm_split = SplitClassifier(LinearCSVMC())<br>
rfeclf = \<br>
FeatureSelectionClassifier(<br>
clf = LinearCSVMC(),<br>
feature_selection = RFE(<br>
sensitivity_analyzer = rfesvm_split.get_sensitivity_analyzer(),<br>
transfer_error = ConfusionBasedError(<br>
rfesvm_split,<br>
confusion_state="confusion"),<br>
feature_selector = FractionTailSelector(<br>
0.2, mode='discard', tail='lower'),<br>
update_sensitivity=True),<br>
enable_states=['feature_ids'],<br>
descr='LinSVM+RFE(splits_avg)')<br>
<br>The error:<br>Traceback (most recent call last):<br> File "<stdin>", line 8, in <module><br> File "/usr/lib/pymodules/python2.6/mvpa/clfs/transerror.py", line 1500, in __init__<br> (confusion_state, clf)<br>
ValueError: Conditional attribute confusion is not defined for classifier SplitClassifier(combiner=None, space='targets')<br><br>"confusion" is the ID of the conditional attribute that stores ConfusionMatrix, but since confusion matrix isn't computed until data is acquired (as stated from <a href="http://pymvpa.org/generated/mvpa.clfs.transerror.ConfusionMatrix.html#mvpa-clfs-transerror-confusionmatrix">http://pymvpa.org/generated/mvpa.clfs.transerror.ConfusionMatrix.html#mvpa-clfs-transerror-confusionmatrix</a>), am I missing something before this code that I should have and that is why it gives that error?<br>
<br>Or must I enable the conditional attribute confusion in some way before this?<br><br>I would really appreciate any Ideas or input as we are pretty stuck at this point...<br>and of course, thanks in advance for any replies!<br>
<br>Kimberly<br><br><br>
<br><br>