spark-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Feynman Liang <fli...@databricks.com>
Subject Re: Label based MLLib MulticlassMetrics is buggy
Date Wed, 05 Aug 2015 20:16:21 GMT
1.5 has not yet been released; what is the commit hash that you are
building?

On Wed, Aug 5, 2015 at 10:29 AM, Hayri Volkan Agun <volkanagun@gmail.com>
wrote:

> Hi,
>
> In Spark 1.5 I saw a result for precision 1.0 and recall 0.01 for decision
> tree classification.
> While precision a hundred the recall shouldn't be so small...I checked the
> code, everything seems ok
> but why I got such a result is unexplainable. As far as I understand from
> scala code the row sum is the actual
> class counts, the column sum is predictions sum am I right?
> I am doing additional tests for comparison with my own code...
> I attached a document for my reuters tests on page 3.
>
>
> On Wed, Aug 5, 2015 at 7:57 PM, Feynman Liang <fliang@databricks.com>
> wrote:
>
>> Also, what version of Spark are you using?
>>
>> On Wed, Aug 5, 2015 at 9:57 AM, Feynman Liang <fliang@databricks.com>
>> wrote:
>>
>>> Hi Hayri,
>>>
>>> Can you provide a sample of the expected and actual results?
>>>
>>> Feynman
>>>
>>> On Wed, Aug 5, 2015 at 6:19 AM, Hayri Volkan Agun <volkanagun@gmail.com>
>>> wrote:
>>>
>>>> The results in MulticlassMetrics is totally wrong. They are improperly
>>>> calculated.
>>>> Confusion matrix may be true I don't know but for each label scores are
>>>> wrong.
>>>>
>>>> --
>>>> Hayri Volkan Agun
>>>> PhD. Student - Anadolu University
>>>>
>>>
>>>
>>
>
>
> --
> Hayri Volkan Agun
> PhD. Student - Anadolu University
>

Mime
View raw message