Apache SystemML 2016 Summer class primer by Berthold Reinwald
- 2. Class Description • Goal • Teach scalable machine learning with Apache SystemML • Attract potential contributors • Audience • Initially summer interns, but goal of developing / folding into University class • Duration ~16 hours • Content • Development of scalable machine learning algorithms • SystemML usage and hands-on exercises • Advanced SystemMLinternals • Office hours • At Adlab: Thursday, 4-5 pm (may be expanded as demanded) 2
- 3. Outline 1. SystemML Primer 2. Machine Learning Algorithms 3. Advanced SystemML Internals 3
- 4. SystemML Primer • Goal • Teach enough DML, SystemML usage, and Spark for people to be able to write and run SystemML algorithms on Spark and understand its execution. • Content • DML syntax • SystemML usage • Some Spark 4
- 5. Machine Learning Algorithms • Descriptive Statistics, Data Preparation, and Train/Test/Cross-Validation • Regression • Classification • Clustering & Matrix Factorization 5 For each session / chosen algorithm have a similar structure: • Possible Applications • Math / Alternatives / Discussion • DML formulation • Data generation • Hands-on exercises • Performance • Accuracy
- 6. Advanced SystemML Internals • Architecture • Compiler • Rewrites • Optimizer • Runtime • Buffer pool • Storage • Advanced Operators • Spark Backend • Performance debugging 6
- 7. 7 S# / Date Category Title Content Instructor S# / Date Category Title Content Instructor S1 6/21: 9-12 am R: G1-404 SystemML Primer Scalable Machine Learning with Apache SystemML • Intro ML • DML • SystemML usage • Architecture Berthold Reinwald, Nakul Jindal S5 7/18 4-6 R: ML Algs Clustering & Matrix Factorization • kMeans, mf, ALS, PCA, …) • DML • Data gen • Hands-on • Perf & Accuracy Alexandre Evfimievski, Prithvi Sen S2 6/27: 4-6 pm R: ML Algs Data Prep, Descriptive Statistics, and Train/Test/Cr oss-validation • Math • DML • Data-gen • Hands-on • Perf & Accuracy Faraz Makari Manshadi S6 7/25 4-6 pm R: SystemML Internals Apache SystemML Architect. • Architecture • Hops/Lops • CP/Cluster Berthold Reinwald, Niketan Pansare S3 7/5: 4-6 R: ML Algs Regression • Linear, log., GLM, Cox, Time series; CG method • DML • Data-gen • Hands-on • Perf & Accuracy Alexandre Evfimievski S7 8/1 4-6 pm R: SystemML Internals Apache SystemML Optimizer • Rewrites • Optimizer • Cost model Matthias Boehm, Arvind Surve S4 7/11: 4-6 R: ML Algs Classificat. • NaïveBayes, SVM, decTree, RF • DML • Data-gen • Hands-on • Perf & Accuracy Prithvi Sen S8 8/8 4-6 pm R: SystemML Internals Apache SystemML Runtime • Buffer pool • Storage • Spark backend • Matrix block lib • Performance debugging Matthias Boehm, Arvind Surve