H2O in KNIME: Integrating High Performance Machine Learning

Size: px

Start display at page:

Download "H2O in KNIME: Integrating High Performance Machine Learning"

Gillian Ross
6 years ago
Views:

3 3

4 Gartner 2018 Magic Quadrant for Data Science and Machine Learning Platforms 4

Supervised & Unsupervised ing Evaluation & Selection Predict Data & Storage Local SQL Data Prep

5 High Level Architecture HDFS H 2 O Compute Engine S3 NFS Load Data Distributed In-Memory Loss-less Compression Exploratory & Descriptive Analysis Feature Engineering & Selection Supervised & Unsupervised ing Evaluation & Selection Predict Data & Storage Local SQL Data Prep Export: Plain Old Java Object Export: Plain Old Java Object Production Scoring Environment Your Imagination 5

6 High Level Architecture Multiple Interfaces HDFS H 2 O Compute Engine S3 NFS Load Data Distributed In-Memory Loss-less Compression Exploratory & Descriptive Analysis Feature Engineering & Selection Supervised & Unsupervised ing Evaluation & Selection Predict Data & Storage Local SQL Data Prep Export: Plain Old Java Object Export: Plain Old Java Object Production Scoring Environment Your Imagination 6

7 Algorithms Overview Supervised Learning Unsupervised Learning Statistical Analysis Generalized Linear s: Binomial, Gaussian, Gamma, Poisson and Tweedie Naïve Bayes Clustering K-means: Partitions observations into k clusters/groups of the same spatial size. Automatically detect optimal k Ensembles Distributed Random Forest: Classification or regression models Gradient Boosting Machine: Produces an ensemble of decision trees with increasing refined approximations Dimensionality Reduction Principal Component Analysis: Linearly transforms correlated variables to independent components Generalized Low Rank s: extend the idea of PCA to handle arbitrary data consisting of numerical, Boolean, categorical, and missing data Deep Neural Networks Deep learning: Create multi-layer feed forward neural networks starting with an input layer followed by multiple layers of nonlinear transformations Anomaly Detection Autoencoders: Find outliers using a nonlinear dimensionality reduction using deep learning 7

8 Scientific Advisory Council 8

9 H 2 O on a Single Machine Building CPU H2O

10 H 2 O on a Multi-Node Cluster Firewall or Cloud Building SQL NFS H2O Distributed In-Memory S3 CPU CPU CPU YARN

11 Foundation for Distributed Algorithms Distributed Algorithms Parallel Parse into Distributed Rows Fine Grain Map Reduce Illustration: Scalable Distributed Histogram Calculation for GBM Advantageous Foundation Foundation for In-Memory Distributed Algorithm Calculation - Distributed Data Frames and columnar compression All algorithms are distributed in H 2 O: GBM, GLM, DRF, Deep Learning and more. Fine-grained map-reduce iterations. Only enterprise-grade, open-source distributed algorithms in the market User Benefits Out-of-box functionalities for all algorithms (NO MORE SCRIPTING) and uniform interface across all languages: R, Python, Java Designed for all sizes of data sets, especially large data Highly optimized Java code for model exports In-house expertise for all algorithms 11

Performance Benchmark (GBM) by Szilard Pafka (https://github.

12 Performance Benchmark (GBM) by Szilard Pafka ( Time (s) H 2 O Gradient Boosting Machine (GBM) is the fastest algorithm in the test of 10M Samples. Area Under Curve (AUC) Comparable with xgboost Better than R/Spark MLLib AUC Benchmark results for other algorithms (Generalized Linear, Random Forest etc.) are available in Szilard s GitHub repository. X = million of samples 12

14 H2O in KNIME Offer our users high-performance machine learning algorithms from H2O in KNIME Allow to mix & match with other KNIME functionality Data wrangling KNIME Analytics Platform functionality KNIME Big-Data Connectors Text Mining, Image Processing, Cheminformatics, and more! 2018 KNIME AG. All Rights Reserved. 14

The competition Date Store ID Visitors Provided data: Number of visitors Reservations Store information Calendar date info 2016-01-01 ba937bf13d40fb24

18 The competition Date Store ID Visitors Provided data: Number of visitors Reservations Store information Calendar date info ba937bf13d40fb f7c39a8410e7c 216 Date Store ID Visitors e8ed9335d0c38333? f13ef0f5e8c64dd? 2018 KNIME AG. All Rights Reserved. 18

22 The blended dataset Data used for model training Date Store ID Visitors Mean visitors Area Reservations Holiday Genre ba937bf13d40fb Area1 15 False Bar f7c39a8410e7c Area2 22 True Karaoke Data used for Kaggle prediction Date Store ID Visitors Mean visitors Area Reservations Holiday Genre e8ed9335d0c38333? 12.5 Area1 10 False Italian f13ef0f5e8c64dd? 54.1 Area7 20 False Izakaya 2018 KNIME AG. All Rights Reserved. 22

Solving a Kaggle competition with KNIME and H2O Data preparation training

39 2018 KNIME AG. All Rights Reserved. The KNIME trademark and logo and OPEN FOR INNOVATION trademark are used by KNIME AG under license from KNIME GmbH, and are registered in the United States. KNIME is also registered in Germany. 39

Fast Analytics on Big Data with H20

Fast Analytics on Big Data with H20 0xdata.com, h2o.ai Tomas Nykodym, Petr Maj Team About H2O and 0xdata H2O is a platform for distributed in memory predictive analytics and machine learning Pure Java,