spark-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Feynman Liang <fli...@databricks.com>
Subject Re: RandomForest evaluator for grid search
Date Mon, 13 Jul 2015 21:08:41 GMT
Joseph may be already working on it; I would continue the discussion on
JIRA and first ask if anyone is working on it before starting your own PR.

On Mon, Jul 13, 2015 at 1:47 PM, Olivier Girardot <
o.girardot@lateral-thoughts.com> wrote:

> Thanks ! that's great !
> any way to help on that ?
>
> 2015-07-13 22:39 GMT+02:00 Feynman Liang <fliang@databricks.com>:
>
>> That is currently tracked by SPARK-3727
>> <https://issues.apache.org/jira/browse/SPARK-3727>.
>>
>> On Mon, Jul 13, 2015 at 1:16 PM, Olivier Girardot <
>> o.girardot@lateral-thoughts.com> wrote:
>>
>>> thx for the info.
>>>
>>> I'd be interested in getting the full predict_proba like in scikit learn
>>> (
>>> http://scikit-learn.org/stable/modules/generated/sklearn.ensemble.RandomForestClassifier.html#sklearn.ensemble.RandomForestClassifier.predict_proba)
>>> for the random forest model.
>>> There doesn't seem to be a way to get the details, is there any reason
>>> for that ?
>>>
>>> Regards,
>>>
>>> Olivier.
>>>
>>> Le lun. 13 juil. 2015 à 21:12, Feynman Liang <fliang@databricks.com> a
>>> écrit :
>>>
>>>> There is MulticlassMetrics in MLlib; unfortunately a pipelined version
>>>> hasn't yet been made for spark-ml. SPARK-7690
>>>> <https://issues.apache.org/jira/browse/SPARK-7690> is tracking work
on
>>>> this if you are interested in following the development.
>>>>
>>>> On Mon, Jul 13, 2015 at 2:16 AM, Olivier Girardot <
>>>> o.girardot@lateral-thoughts.com> wrote:
>>>>
>>>>> Hi everyone,
>>>>> Using spark-ml there seems to be only BinaryClassificationEvaluator
>>>>> and RegressionEvaluator, is there any way or plan to provide a ROC-based
or
>>>>> PR-based or F-Measure based for multi-class, I would be interested
>>>>> especially in evaluating and doing a grid search for a RandomForest model.
>>>>>
>>>>> Regards,
>>>>>
>>>>> Olivier.
>>>>>
>>>>
>>>>
>>
>

Mime
View raw message