You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@marvin.apache.org by "Bruno Sette (Jira)" <ji...@apache.org> on 2019/11/26 15:22:00 UTC
[jira] [Created] (MARVIN-70) Keras predictor preparator (image)
nparray is not JSON serializable problem
Bruno Sette created MARVIN-70:
---------------------------------
Summary: Keras predictor preparator (image) nparray is not JSON serializable problem
Key: MARVIN-70
URL: https://issues.apache.org/jira/browse/MARVIN-70
Project: Apache Marvin
Issue Type: Bug
Reporter: Bruno Sette
Hi guys,
I'm having a problem with predictor preparator in dryrun. Keras predictor needs to convert images to nparray, a type wich its not json serializable. I've tried to convert the nparray to a python list, but this doesn't work well because it is a multidimensional array. Any ideas on how to solve this?
Thanks.
{code:java}
------------------------------------------------------------------------------
MARVIN DRYRUN - STEP [PredictionPreparator]
------------------------------------------------------------------------------
Message 0 :
"http://datawrangling.s3.amazonaws.com/sample_digit.png"
Result for Message 0 :
Unable to serialize the object returned!
{code}
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
Re: [jira] [Created] (MARVIN-70) Keras predictor preparator (image)
nparray is not JSON serializable problem
Posted by Daniel Takabayashi <da...@gmail.com>.
Hi Bruno, to solve this problem take a look in the public engines here:
Custom serializer definition:
https://github.com/marvin-ai/marvin-public-engines/blob/master/image-classification-engine/marvin_image_classification_engine/model_serializer.py
Usage of the custom serializer:
https://github.com/marvin-ai/marvin-public-engines/blob/master/image-classification-engine/marvin_image_classification_engine/training/trainer.py
https://github.com/marvin-ai/marvin-public-engines/blob/master/image-classification-engine/marvin_image_classification_engine/prediction/predictor.py
Basically, Marvin gives you the power of creating your own serializable, in
this case, I am just using Keras default serializer.
By default we already have a Keras serializer inside de tools box, you just
to have to use it:
Here is the serializer:
https://github.com/marvin-ai/marvin-python-toolbox/blob/master/marvin_python_toolbox/engine_base/serializers/keras_serializer.py
You just have to extend it from your Predictor and Trainer classes to use
it.
Em ter., 26 de nov. de 2019 às 07:22, Bruno Sette (Jira) <ji...@apache.org>
escreveu:
> Bruno Sette created MARVIN-70:
> ---------------------------------
>
> Summary: Keras predictor preparator (image) nparray is not
> JSON serializable problem
> Key: MARVIN-70
> URL: https://issues.apache.org/jira/browse/MARVIN-70
> Project: Apache Marvin
> Issue Type: Bug
> Reporter: Bruno Sette
>
>
> Hi guys,
>
> I'm having a problem with predictor preparator in dryrun. Keras predictor
> needs to convert images to nparray, a type wich its not json serializable.
> I've tried to convert the nparray to a python list, but this doesn't work
> well because it is a multidimensional array. Any ideas on how to solve this?
>
>
>
> Thanks.
> {code:java}
>
>
> ------------------------------------------------------------------------------
> MARVIN DRYRUN - STEP [PredictionPreparator]
>
> ------------------------------------------------------------------------------
>
> Message 0 :
> "http://datawrangling.s3.amazonaws.com/sample_digit.png"
> Result for Message 0 :
> Unable to serialize the object returned!
> {code}
>
>
>
>
>
>
>
>
>
>
> --
> This message was sent by Atlassian Jira
> (v8.3.4#803005)
>