Intro to AutoML + Hands-on Lab - Erin LeDell, Machine Learning Scientist, H2O.ai
10 likes5,415 views
The document provides an overview of Automatic Machine Learning (AutoML) and its implementation using H2O, detailing aspects such as data preparation, model generation, and ensemble techniques. It outlines a hands-on tutorial agenda, various model training methods, and examples for R and Python, as well as a roadmap for future feature upgrades. Key highlights include the use of random grid search, stacked ensembles, and the ease of integrating H2O with different platforms.
Intro to AutoML + Hands-on Lab - Erin LeDell, Machine Learning Scientist, H2O.ai
- 2. Machine Learning Scientist @ledell / [email protected] Erin LeDell
- 4. MEET THE MAKERS ERIN LEDELL Machine Learning Scientist NAVDEEP GILL Software Engineer & Data Scientist RAY PECK Director of Product Engineering
- 5. • Intro to Automatic Machine Learning (AutoML) • Random Grid Search & Stacked Ensembles • H2O’s AutoML (R, Python, GUI) • H2O-3 Roadmap • Hands-on Tutorial Agenda
- 6. Intro to AutoML Automatic Machine Learning
- 7. Aspects of Automatic ML Model Generation EnsemblesData Prep
- 8. Data Prep • Imputation of missing data • Standardization of numeric features • One-hot encoding of categorical features • Count/Label/Target encoding of categorical features • Feature selection and/or feature extraction (e.g. PCA) • Feature engineering
- 9. Model Generation • Cartesian grid search • Random grid search • Tune individual models via Early Stopping • Bayesian Hyperparameter Optimization
- 10. Ensembles • Bagging / Averaging • Stacking / Super Learning • Ensemble Selection
- 11. Random Stacking Random Grids + Stacked Ensembles
- 12. Stacked Ensembles • Specify L base learners (with model params). • Specify a metalearner (just another algo). • Perform k-fold cross-validation on the base learners.
- 13. Stacked Ensembles • Collect cross-validated predicted values from base learners. • Train a second-level metalearning algorithm to find the optimal combination of base learners. • Metalearner requires only a small amount of compute on top of the cross-validation process (it’s cheap).
- 14. Random Grid Search + Stacking • Random Grid Search combined with Stacked Ensembles is a powerful combination. • Stacked Ensembles perform particularly well if the models they are based on (1) are individually strong, and (2) make uncorrelated errors. • Random Grid Search is an excellent way to create a diverse of models for the ensemble.
- 15. H2O AutoML Automatic Machine Learning in H2O
- 16. 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. • Works on Hadoop, Spark, EC2, your laptop, etc. • Easily deploy models to production as pure Java code.
- 17. H2O AutoML (first cut) • Imputation, one-hot encoding, standardization. • Random Grid Search over a custom hyperparameter space, defined by expert data scientists. • Early stopping of individual models and random grids. • GBMs, Random Forests, Deep Neural Nets, GLMs • Multiple Stacked Ensembles of models. • Leaderboard for ranking.
- 18. H2O AutoML in R library(h2o) h2o.init() train <- h2o.importFile("train.csv") aml <- h2o.automl(y = "response_colname", training_frame = train, max_runtime_secs = 600) lb <- aml@leaderboard
- 19. H2O AutoML in Python import h2o from h2o.automl import H2OAutoML h2o.init() train = h2o.import_file("train.csv") aml = H2OAutoML(max_runtime_secs = 600) aml.train(y = "response_colname", training_frame = train) lb = aml.leaderboard
- 20. H2O AutoML in Flow
- 21. Example Leaderboard for binary classification H2O AutoML Leaderboard
- 22. H2O-3 Roadmap Coming Soon to H2O
- 23. Feature Q1 Q2 New Algorithm: Cox-Proportional Hazards GLM: Ordinal Regression GBM: Quasibinomial NLP Improvements, TF-IDF Stacked Ensemble: Custom Metalearner AutoML: New Ensembles AutoML: Add XGBoost Distributed XGBoost New Algorithm: Factorization Machines H2O-3 Roadmap https://tinyurl.com/h2o-automl-jira
- 24. • 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 • Events & Meetups: http://h2o.ai/events • Stack Overflow: https://stackoverflow.com/tags/h2o • Google Group: https://tinyurl.com/h2ostream • Gitter: http://gitter.im/h2oai/h2o-3
- 26. First-time Qwiklab Account Setup • Go to http://h2oai.qwiklab.com • Click on “JOIN” • Create a new account with a valid email address • You will receive a confirmation email • Click on the link in the confirmation email • Go back to http://h2oai.qwiklab.com and log in • Go to the Catalog on the left bar • Choose “Introduction to AutoML in H2O” • Wait for instructions
- 27. https://tinyurl.com/automl-h2oworld17 Code and data available here H2O AutoML Tutorial