<div dir="ltr"><div>I've trained a GNB classifier on a dataset with moderate success and would like to look at the regions that the classifier finds most interesting. I notice in the code that there is a sensitivity analyser for the GNB module but that it has been removed:<br><br><table class=""><tbody><tr><td id="LC293" class=""><span class=""># XXX Later come up with some</span></td>
      </tr>
      <tr>
        
        <td id="LC294" class="">    <span class="">#     could be a simple t-test maps using distributions</span></td>
      </tr>
      <tr>
        
        <td id="LC295" class="">    <span class="">#     per each class</span></td>
      </tr>
      <tr>
        
        <td id="LC296" class="">    <span class="">#def get_sensitivity_analyzer(self, **kwargs):</span></td>
      </tr>
      <tr>
        
        <td id="LC297" class="">    <span class="">#    """Returns a sensitivity analyzer for GNB."""</span></td>
      </tr>
      <tr>
        
        <td id="LC298" class="">    <span class="">#    return GNBWeights(self, **kwargs)</span></td>
      </tr>
      <tr>
        
        <td id="LC299" class="">
<br></td>
      </tr>
      <tr>
        
        <td id="LC300" class="">
<br></td>
      </tr>
      <tr>
        
        <td id="LC301" class="">    <span class=""># XXX Is there any reason to use properties?</span></td>
      </tr>
      <tr>
        
        <td id="LC302" class="">    <span class="">#means = property(lambda self: self.__biases)</span></td>
      </tr>
      <tr>
        
        <td id="LC303" class="">    <span class="">#variances = property(lambda self: self.__weights)</span></td></tr></tbody></table><br><br><table class=""><tbody><tr><td id="LC307" class=""><span class="">## class GNBWeights(Sensitivity):</span></td>
      </tr>
      <tr>
        
        <td id="LC308" class=""><span class="">##     """`SensitivityAnalyzer` that reports the weights GNB trained</span></td>
      </tr>
      <tr>
        
        <td id="LC309" class=""><span class="">##     on a given `Dataset`.</span></td>
      </tr>
      <tr>
        
        <td id="LC310" class="">
<br></td>
      </tr>
      <tr>
        
        <td id="LC311" class=""><span class="">##     """</span></td>
      </tr>
      <tr>
        
        <td id="LC312" class="">
<br></td>
      </tr>
      <tr>
        
        <td id="LC313" class=""><span class="">##     _LEGAL_CLFS = [ GNB ]</span></td>
      </tr>
      <tr>
        
        <td id="LC314" class="">
<br></td>
      </tr>
      <tr>
        
        <td id="LC315" class=""><span class="">##     def _call(self, dataset=None):</span></td>
      </tr>
      <tr>
        
        <td id="LC316" class=""><span class="">##         """Extract weights from GNB classifier.</span></td>
      </tr>
      <tr>
        
        <td id="LC317" class="">
<br></td>
      </tr>
      <tr>
        
        <td id="LC318" class=""><span class="">##         GNB always has weights available, so nothing has to be computed here.</span></td>
      </tr>
      <tr>
        
        <td id="LC319" class=""><span class="">##         """</span></td>
      </tr>
      <tr>
        
        <td id="LC320" class=""><span class="">##         clf = self.clf</span></td>
      </tr>
      <tr>
        
        <td id="LC321" class=""><span class="">##         means = clf.means</span></td>
      </tr>
      <tr>
        
        <td id="LC322" class=""><span class="">##           XXX we can do something better ;)</span></td>
      </tr>
      <tr>
        
        <td id="LC323" class=""><span class="">##         return mean</span></td></tr></tbody></table><br></div><div>Is the use of the means considered to be poor in some sense?<br></div><div><br></div><div>Could anyone provide more information about this:<br><br><table class=""><tbody><tr><td id="LC293" class=""><span class=""># XXX Later come up with some</span></td></tr><tr><td id="LC294" class="">    <span class="">#     could be a simple t-test maps using distributions</span></td></tr><tr><td id="LC295" class="">    <span class="">#     per each class</span></td></tr></tbody></table><br></div><div><br></div>Thanks,<br><br>Tom<br></div>