[SPARK-10478] Improve spark.ml.ann implementations for MLP

• Type: Improvement
• Status: Resolved
• Priority: Major
• Resolution: Incomplete
• Component/s: ML
• Labels: bulk-closed
• Assignee: Unassigned
• Reporter: Feynman Liang
• Created: 07/Sep/15 21:41
• Updated: 21/May/19 04:37
• Resolved: 21/May/19 04:37

Description

SPARK-9471 adds an implementation of multi-layer perceptrons. However, there are a few issues with the current code that should be addressed, namely:

• Style guide: 4 indent method arguments, braces around one-line {{if}}s, punctuation in scaladocs
• Comments: cryptic variable names (e.g. gw, gb, gwb) should be documented
• Performance: parts of the code can be improved using Breeze broadcasts, vectorized operations, and Breeze UFuncs to avoid manual iteration and leverage BLAS optimization

Issue Links

• is related to: SPARK-9471 Multilayer perceptron classifier
• relates to: SPARK-11262 Unit test for gradient, loss layers, memory management for multilayer perceptron
• links to: [Github] Pull Request #8648 (feynmanliang)