Key takeaways- H2O co-founders Srisatish Ambati and Cliff Click presented a talk at Strata NYC and Hadoop World on modeling big data with better algorithms using H2O.
- The talk description said H2O is an open source math and machine learning engine for big data that brings distribution and parallelism while keeping R and JSON as its API.
- The talk covered design choices behind H2O's Distributed Random Forest and Generalized Linear Modeling, plus a hacking demo showing data manipulation, summarization, modeling, grid search, and scoring.
How to Stop Worrying and Start Modeling Big Data with Better Algorithms and H2O Srisatish Ambati (0xdata Inc), Cliff Click (0xdata Inc) 5:05pm Tuesday, 10/29/2013 Data Science Beekman Parlor – Sutton North Data Modeling has been constrained through scale; Sampling still rules the day for Adhoc Analytics. Scale brings much needed change to the modeling world. In this talk we present the predictive power of using sophisticated algorithms on big datasets. With large data sizes comes the particularly hard problem of unbalanced data with multiple asymmetrically rare classes. Missing features pose unique problems for most Classification and Regression algorithms and proper handling can lead to greater predictive power. In the race for Better Predictions, H2O makes practical techniques accessible to manyone through an easy-to-use software product. H2O is an open source math & machine learning engine for big data that brings distribution and parallelism to powerful algorithms while keeping the widely used languages of R and JSON as an API. And integrates neatly into popular data ecosystems of hadoop, amazon s3, nosql and sql. We briefly discuss design choices in the implementation of Distributed Random Forest and Generalized Linear Modeling and bringing speed and scale to vox populi of Data Science, R. We take a peek at the elegant lego-like infrastructure that brings fine grained parallelism to math over simple distributed arrays. A short hacking data demo presents the life cycle of Data Science: Powerful Data Manipulation via R at scale, Interactive Summarization over large datasets, Modeling using Elastic Net (GLM), Grid Search for best parameters & low-latency scoring.
:quality(75))