Apache Hivemall 0.5.2 发布，可扩展的机器学习库

  

Apache Hivemall 0.5.2 发布了，Apache Hivemall 基于 Hive UDF/UDAF/UDTF，是一个可扩展的机器学习库，运行基于 Hadoop 的数据处理框架，特别是 Apache Hive、Apache Spark 和 Apache Pig。

更新主要内容包括：

New Feature

• [HIVEMALL-145] - Merge brickhouse functions

Improvement

• [HIVEMALL-24] - Fix the prediction logic of Field-aware Factorization Machines more scalable

• [HIVEMALL-46] - Make it more simpler to upgrade Spark versions

• [HIVEMALL-172] - Change tree_predict 3rd argument to accept string options

• [HIVEMALL-179] - Support Spark 2.3

• [HIVEMALL-180] - Drop the Spark-2.0 support

• [HIVEMALL-191] - Add Kryo serialization tests and remove existing workaround lazy instantiation code

• [HIVEMALL-193] - Implement a tool for generating a list of Hivemall UDFs

• [HIVEMALL-201] - Evaluate, fix and document FFM so Hivemall produces comparable accuracy to LIBFFM

• [HIVEMALL-203] - Relocate Jackson package for to_json/from_json

• [HIVEMALL-212] - Fix Classifier/Regressor not to forward zero weighted values

• [HIVEMALL-215] - [DOC] Add step-by-step tutorial on the document

• [HIVEMALL-222] - Introduce Gradient Clipping to avoid exploding gradient to General Classifier/Regressor

• [HIVEMALL-223] - Add `-kv_map` and `-vk_map` option to to_ordered_list UDAF

详情查看更新日志。

下载地址：http://hivemall.incubator.apache.org/download.html