train_arowe_regr


train_arowe_regr(array features, float target [, constant string options]) – a refined version of AROW (Adaptive Reguralization of Weight Vectors) regressor that usages epsilon-insensitive hinge loss `|w^t – y| – epsilon` for the loss function

SELECT
feature,
argmin_kld(weight, covar) as weight
FROM (
SELECT
train_arowe_regr(features,label) as (feature,weight,covar)
FROM
training_data
) t
GROUP BY feature

Platforms: WhereOS, Spark, Hive
Class: hivemall.regression.AROWRegressionUDTF$AROWe

More functions can be added to WhereOS via Python or R bindings or as Java & Scala UDF (user-defined function), UDAF (user-defined aggregation function) and UDTF (user-defined table generating function) extensions. Custom libraries can be added on via Settings-page or installed from WhereOS Store.

Leave a Reply