Andrew Musselman, Committer & PMC Member, Apache Mahout at MLconf Seattle 2017
summary
Apache Mahout 0.13 provides a Scala R-like DSL (Samsara) that compiles linear-algebra expressions to multiple backends (Apache Spark, Apache Flink, H2O) and an Apache Zeppelin interpreter for interactive notebooks. It ships a hybrid distributed/GPU-accelerated engine with native GPU work and JCUDA bindings, single-GPU runs verified, and multi-GPU support next. An scikit-learn/CRAN-style algorithm framework accepts user PRs. Small benchmarks show about a 30x speedup on 1000×1000 matrix multiply versus in-core execution.
AI-written summary · about
chapters (18)
- 0:00Introduction
- 0:37About Andrew
- 1:03Recent developments
- 2:04Highlevel view
- 2:48Timeline
- 4:26Overview
- 4:38Samsara
- 5:13DSL
- 7:44Algebraic Optimization
- 9:07Backend
- 9:24Zeppelin Notebook
- 11:08zeppelin
- 11:28spark
- 11:52scikitlearn
- 13:24Benchmarking
- 14:03Next Steps
- 14:23Give Away
- 14:34Dev List
related talks
30mApache SystemML: State of the Project and Future Plans
39mMachine Learning with Scala on Spark
16mAmeet Talwalkar, Assistant Professor of Computer Science, UCLA @ MLconf SF Part 2
22mJuliet Hougland, Data Scientist, Cloudera
51mApache Spark and machine learning: dreams and realities
38mFantastic ML apps and how to build them
24mJoseph K. Bradley, Software Engineer, Databricks Inc. @ MLconf SEA
50mApache Spark vs Machine Learning