Mike Seddon created SPARK-13097:
-----------------------------------
Summary: Extend Binarizer to allow Double AND Vector inputs
Key: SPARK-13097
URL: https://issues.apache.org/jira/browse/SPARK-13097
Project: Spark
Issue Type: Improvement
Components: ML
Reporter: Mike Seddon
Priority: Minor
To enhance the existing SparkML Binarizer [SPARK-5891] to allow Vector in
addition to the existing Double input column type.
A use case for this enhancement is for when a user wants to Binarize many
similar feature columns at once using the same threshold value.
A real-world example for this would be where the authors of one of the leading
MNIST handwriting character recognition entries converts 784 grayscale (0-255)
pixels (28x28 pixel images) to binary if the pixel's grayscale exceeds 127.5:
(http://arxiv.org/abs/1003.0358). With this modification the user is able to:
VectorAssembler->Binarizer(127.5)->Classifier as all the pixels are of a
similar type.
This approach also allows much easier use of the ParamGridBuilder to test
multiple theshold values.
I have already written the code and unit tests and have tested in a Multilayer
perceptron classifier workflow.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org
For additional commands, e-mail: issues-h...@spark.apache.org
