A team member came up with a clever solution. For anyone's reference, one
way is to place the Gradient Boosting Decision Tree classification part
(via XGBoost) in a separate Python process, to get around logging that can
raise errors. Then the Storm bolt can communicate with that via a
multiprocessing library. Related links:

   - http://stackoverflow.com/questions/6920858/
   interprocess-communication-in-python
   
<http://stackoverflow.com/questions/6920858/interprocess-communication-in-python>
   - https://docs.python.org/3/library/multiprocessing.html

On Wed, Apr 12, 2017 at 1:20 AM, Derek S. Chan <[email protected]> wrote:

> Hi All,
>
> I'm part of a graduate team at UC Berkeley. In Storm, does anyone know
> how users can silence logging from XGBoost's predict (not train) method? My
> team has been delayed for several days, unclear how to get around this
> logging issue, and seeks to overcome ASAP.
>
> Thanks,
> Derek
>

Reply via email to