mahout-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Michael Christopher <mi...@bluewin.ch>
Subject Re: Calculation of Kappa value
Date Sun, 25 May 2014 16:21:15 GMT
Thanks Sebastian

No, I downloaded the source code (version 0.9) from mirror. I am bit 
busy at the moment. will try to work with the trunk and provide a patch 
in about 2 weeks.


Regards

Michael


On 25.05.2014 17:08, Sebastian Schelter wrote:
> Could be a bug introduced by a recent modification of the Confusion 
> matrix. Are you using trunk? Can you provide a patch that fixes the 
> issue?
>
> Best,
> Sebastian
>
>
> On 05/25/2014 05:05 PM, Michael Christopher wrote:
>> Hi
>>
>> With the Confusion Matrix below I get a Kappa value of 0,8023. Actually
>> it should be 1, because there are no wrong classifications.
>>
>> I debugged and found out that it happens, because the samples variable
>> is two times incremented in the "putCount" method
>> (org.apache.mahout.classifier.ConfusionMatrix). Commenting out this
>> incrementation lead to the Kappa value of 1.
>>
>> Is this a bug or do I something something?
>>
>>
>> =======================================================
>> Confusion Matrix
>> -------------------------------------------------------
>> a        b        <--Classified as
>> 6        0         |  6         a     = 1_YES
>> 0        9         |  9         b     = 2_NO
>>
>> =======================================================
>> Statistics
>> -------------------------------------------------------
>> Kappa                                       0,8023
>> Accuracy                                       100%
>> Reliability                                66,6667%
>> Reliability (standard deviation)            0,5774
>>
>>
>>
>>
>>
>


Mime
View raw message