Open Platform for AI & ML modeling
1 like80 views
The document presents an overview of H2O.ai and its machine learning platform H2O, which includes features like automatic machine learning (AutoML) to streamline model training and enhance performance. H2O supports various programming languages and provides APIs for easy deployment, with emphasis on distributed computing and ensemble methods for improved accuracy. The text also offers practical tips for utilizing AutoML and mentions future roadmap developments.
Open Platform for AI & ML modeling
- 1. Scalable Automatic Machine Learning in H2O Erin LeDell Ph.D. @ledell Data Science Conference 5.0 Belgrade, Serbia Nov 2019
- 2. What is H2O? H2O.ai, the company H2O, the platform • Founded in 2012 • Advised by Stanford Professors Hastie, Tibshirani & Boyd • Headquarters: Mountain View, California, USA • Open Source Software (Apache 2.0 Licensed) • R, Python, Scala, Java and Web Interfaces • Distributed Machine Learning Algorithms for Big Data
- 3. Agenda • H2O Platform • Automatic Machine Learning (AutoML) • H2O AutoML Overview • Resources Slides ⬇ https://tinyurl.com/automl-dsc5
- 4. H2O Platform
- 5. H2O Machine Learning Platform • Distributed (multi-core + multi-node) implementations of cutting edge ML algorithms. • Core algorithms written in high performance Java. • APIs available in R, Python, Scala; web GUI. • Easily deploy models to production as pure Java code. • Works on Hadoop, Spark, EC2, your laptop, etc.
- 6. 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
- 7. H2O Machine Learning Features • Supervised & unsupervised machine learning algos (GBM, RF, DNN, GLM, Stacked Ensembles, etc.) • Imputation, normalization & auto one-hot-encoding • Automatic early stopping • Cross-validation, grid search & random search • Variable importance, model evaluation metrics, plots
- 8. Intro to Automatic Machine Learning
- 9. Goals & Features of AutoML • 🏆 Train the best model in the least amount of time. • 📉 Reduce the human effort & expertise required in machine learning. • 📈 Improve the performance of machine learning models. • 🔄 Increase reproducibility & establish a baseline for scientific research or applications.
- 10. Aspects of Automatic Machine Learning Data Prep Model Generation Ensembles
- 11. Aspects of Automatic Machine Learning • Cartesian grid search or random grid search • Bayesian Hyperparameter Optimization • Individual models can be tuned using a validation set Data Preprocessing Model Generation Ensembles • Imputation, one-hot encoding, standardization • Feature selection and/or feature extraction (e.g. PCA) • Count/Label/Target encoding of categorical features • Ensembles often out-perform individual models • Stacking / Super Learning (Wolpert, Breiman) • Ensemble Selection (Caruana)
- 12. Different Flavors of AutoML https://tinyurl.com/flavors-of-automl
- 13. H2O’s AutoML
- 14. H2O AutoML (v3.26) • Cartesian grid search or random grid search • Bayesian Hyperparameter Optimization • Individual models can be tuned using a validation set Data Preprocessing Model Generation Ensembles • Imputation, one-hot encoding, standardization • Feature selection and/or feature extraction (e.g. PCA) • Count/Label/Target encoding of categorical features • Ensembles often out-perform individual models: • Stacking / Super Learning (Wolpert, Breiman) • Ensemble Selection (Caruana)
- 15. Random Grid Search & Stacking • Random Grid Search combined with Stacked Ensembles is a powerful combination. • Ensembles perform particularly well if the models they are based on (1) are individually strong, and (2) make uncorrelated errors. • Stacking uses a second-level metalearning algorithm to find the optimal combination of base learners.
- 16. H2O AutoML • Basic data pre-processing (as in all H2O algos). • Trains a random grid of GBMs, DNNs, GLMs, etc. using a carefully chosen hyper-parameter space. • Individual models are tuned using cross-validation. • Two Stacked Ensembles are trained (“All Models” ensemble & a lightweight “Best of Family” ensemble). • Returns a sorted “Leaderboard” of all models. • All models can be easily exported to production.
- 17. H2O AutoML in Python
- 18. H2O AutoML in R
- 19. H2O AutoML in Flow GUI
- 20. H2O AutoML Leaderboard Example Leaderboard for binary classification (Higgs 10k)
- 21. AutoML Pro Tips!
- 22. AutoML Pro Tips: Exclude Algos • If you have sparse, wide data (e.g. text), use the exclude_algos argument to turn off the tree-based models (GBM, RF). • If you want tree-based algos only, turn off GLM and DNNs via exclude_algos.
- 23. AutoML Pro Tips: Time & Model Limits • AutoML will stop after 1 hour unless you change max_runtime_secs. • If you need reproducibility, you must use use max_models instead. Running with max_runtime_secs is not reproducible since available resources on a machine may change from run to run.
- 24. AutoML Pro Tips: Cluster memory • Reminder: All H2O models are stored in H2O Cluster memory. • Make sure to give the H2O Cluster a lot of memory if you’re going to create hundreds or thousands of models. • e.g. h2o.init(max_mem_size = “80G”)
- 25. H2O AutoML Roadmap • Automatic target encoding of high cardinality categorical cols • Better support for wide datasets via feature selection/extraction • Support text input directly via Word2Vec • Variable importance for Stacked Ensembles • Improvements to the models we train based on benchmarking • Fully customizable model list (grid space, etc) • New algorithms: SVM, GAM
- 26. H2O AutoML in the Wild
- 27. AutoML Benchmarks arXiv paper ⬇ https://tinyurl.com/automlbenchmark
- 30. Learn H2O AutoML! • Docs: https://tinyurl.com/h2o-automl-docs • R & Py tutorials: https://tinyurl.com/h2o-automl-tutorials
- 31. H2O Resources • Documentation: http://docs.h2o.ai • Tutorials: https://github.com/h2oai/h2o-tutorials • Slidedecks: https://github.com/h2oai/h2o-meetups • Videos: https://www.youtube.com/user/0xdata • Stack Overflow: https://stackoverflow.com/tags/h2o • Google Group: https://tinyurl.com/h2ostream • Gitter: http://gitter.im/h2oai/h2o-3 • Events & Meetups: http://h2o.ai/events
- 32. Thank you! @ledell on Github, Twitter erin@h2o.ai