Scalable Machine Learning in R and Python with H2O
2 likes2,157 views
This document provides an overview of scalable machine learning in R and Python using the H2O platform. It introduces H2O.ai, the company and H2O, the open source machine learning platform. Key features of the H2O platform include its distributed algorithms, APIs for R and Python, and interfaces like H2O Flow. The document outlines tutorials for using popular algorithms like deep learning and ensembles in H2O and describes ongoing developments like DeepWater and AutoML.
More Related Content
What's hot (20)
Similar to Scalable Machine Learning in R and Python with H2O (20)
More from Sri Ambati (20)
Recently uploaded (20)
Scalable Machine Learning in R and Python with H2O
- 1. Scalable Machine Learning in R and Python with H2O Erin LeDell Ph.D. Machine Learning Scientist H2O.ai Berkeley, CA April 2017 East Bay AI and DL Meetup
- 2. Introduction • Statistician & Machine Learning Scientist at H2O.ai, in Mountain View, California, USA • Ph.D. in Biostatistics with Designated Emphasis in Computational Science and Engineering from UC Berkeley (focus on Machine Learning) • Worked as a data scientist at several startups
- 3. Agenda • Who/What is H2O? • H2O Machine Learning Platform • H2O in R and Python • H2O Tutorials: Intro, Grid, DL, Stacking • New/active developments in H2O
- 4. H2O.ai H2O.ai, the Company H2O, the Platform • Founded in 2012 • Stanford & Purdue Math & Systems Engineers • Headquarters: Mountain View, California, USA • Open Source Software (Apache 2.0 Licensed) • R, Python, Scala, Java and Web Interfaces • Distributed Algorithms that Scale to Big Data
- 5. Scientific Advisory Council • John A. Overdeck Professor of Mathematics, Stanford University • PhD in Statistics, Stanford University • Co-author, The Elements of Statistical Learning: Prediction, Inference and Data Mining • Co-author with John Chambers, Statistical Models in S • Co-author, Generalized Additive Models Dr. Trevor Hastie • Professor of Statistics and Health Research and Policy, Stanford University • PhD in Statistics, Stanford University • Co-author, The Elements of Statistical Learning: Prediction, Inference and Data Mining • Author, Regression Shrinkage and Selection via the Lasso • Co-author, An Introduction to the Bootstrap Dr. Robert Tibshirani • Professor of Electrical Engineering and Computer Science, Stanford University • PhD in Electrical Engineering and Computer Science, UC Berkeley • Co-author, Distributed Optimization and Statistical Learning via the Alternating Direction Method of Multipliers • Co-author, Linear Matrix Inequalities in System and Control Theory • Co-author, Convex Optimization Dr. Steven Boyd
- 6. H2O Platform
- 7. H2O Platform Overview • Distributed implementations of cutting edge ML algorithms. • Core algorithms written in high performance Java. • APIs available in R, Python, Scala, REST/JSON. • Interactive Web GUI called H2O Flow. • Easily deploy models to production with H2O Steam.
- 8. H2O Platform Overview • Write code in high-level language like R (or use the web GUI) and output production-ready models in Java. • To scale, just add nodes to your H2O cluster. • Works with Hadoop, Spark and your laptop.
- 9. H2O Distributed Computing H2O Cluster H2O Frame • Multi-node cluster with shared memory model. • All computations in memory. • Each node sees only some rows of the data. • No limit on cluster size. • Distributed data frames (collection of vectors). • Columns are distributed (across nodes) arrays. • Works just like R’s data.frame or Python Pandas DataFrame
- 10. Current Algorithm Overview Statistical Analysis • Linear Models (GLM) • Naïve Bayes Ensembles • Random Forest • Distributed Trees • Gradient Boosting Machine • Stacked Ensembles Deep Neural Networks • Multi-layer Feed-Forward Neural Network • Auto-encoder • Anomaly Detection • Deep Features Clustering • K-Means Dimension Reduction • Principal Component Analysis • Generalized Low Rank Models Solvers & Optimization • Generalized ADMM Solver • L-BFGS (Quasi Newton Method) • Ordinary Least-Square Solver • Stochastic Gradient Descent Data Munging • Scalable Data Frames • Sort, Slice, Log Transform
- 11. H2O in R & Python
- 12. h2o R Package Installation • Java 7 or later; R 3.1 and above; Linux, Mac, Windows • The easiest way to install the h2o R package is CRAN. • Latest version: http://www.h2o.ai/download/h2o/r Design All computations are performed in highly optimized Java code in the H2O cluster, initiated by REST calls from R.
- 13. h2o Python Module Installation • Java 7 or later; Python 2.7, 3.5; Linux, Mac, Windows • The easiest way to install the h2o Python module is PyPI. • Latest version: http://www.h2o.ai/download/h2o/python Design All computations are performed in highly optimized Java code in the H2O cluster, initiated by REST calls from Python.
- 14. H2O R & Python Tutorials https://github.com/h2oai/h2o-tutorials
- 15. R & Py Tutorial: Intro to H2O Algorithms The “Intro to H2O” tutorial introduces five popular supervised machine learning algorithms in the context of a binary classification problem. The training module demonstrates how to train models and evaluate model performance on a test set. • Generalized Linear Model (GLM) • Random Forest (RF) • Gradient Boosting Machine (GBM) • Deep Learning (DL) • Naive Bayes (NB)
- 16. R & Py Tutorial: Grid Search for Model Selection The second training module demonstrates how to find the best set of model parameters for each model using Grid Search.
- 17. R Tutorial: Deep Learning The “Deep Learning in R” tutorial gives an overview of how to train H2O deep neural networks in R. • Deep Learning via Multilayer Perceptrons (MLPs) • Early Stopping • Random Grid Search • Deep Learning Autoencoders • Unsupervised Pretraining • Deep Features •Anomaly Detection
- 18. Tutorial: Stacked Ensembles There are two H2O Ensemble tutorials: One for the new Stacked Ensemble method in h2o and one for the h2oEnsemble R package which extends the h2o R API.
- 19. H2O Resources • Documentation: http://docs.h2o.ai • Online Training: http://learn.h2o.ai • Tutorials: https://github.com/h2oai/h2o-tutorials • Slidedecks: https://github.com/h2oai/h2o-meetups • Video Presentations: https://www.youtube.com/user/0xdata • Events & Meetups: http://h2o.ai/events
- 21. Thank you! @ledell on Github, Twitter [email protected] http://www.stat.berkeley.edu/~ledell
- 22. Appendix: New & Active Developments in H2O
- 23. Deep Water https://github.com/h2oai/deepwater •Native implementation of Deep Learning models for GPU-optimized backends (mxnet, Caffe, TensorFlow, etc.) •State-of-the-art Deep Learning models trained from the H2O Platform •Provides an easy to use interface to any of the Deep Water backends. •Extends the H2O platform to include Convolutional Neural Nets (CNNs) and Recurrent Neural Nets (RNNs) including LSTMs
- 24. rsparkling https://github.com/h2oai/rsparkling •This provides an interface to H2O's machine learning algorithms on Spark, using R. •This is an extension package for RStudio’s sparklyr package that creates an R front-end for a Spark package (e.g. Sparking Water). •This package implements only the most basic functionality (creating an H2OContext, showing the H2O Flow interface, and converting a Spark DataFrame to an H2OFrame or vice versa).
- 25. H2O AutoML Preview: https://github.com/h2oai/h2o-3/tree/automl •AutoML stands for “Automatic Machine Learning” •The idea here is to remove most (or all) of the parameters from the algorithm, as well as automatically generate derived features that will aid in learning. •Single algorithms are tuned automatically using a carefully constructed random grid search (future: Bayesian Optimization algorithms). •Optionally, a Stacked Ensemble can be constructed.