SlideShare a Scribd company logo

Berlin R Meetup

•
•
Sri Ambati
Sri Ambati

This document summarizes a presentation given by Joe Chow on machine learning using H2O.ai's platform. The presentation introduced H2O, its products like Deep Water for deep learning, and demonstrated examples of building models with R and Python. It showed how H2O provides a unified interface for TensorFlow, MXNet and Caffe, allowing users to easily build and deploy deep learning models with different frameworks. The document provided an overview of the company and platform capabilities like scalable algorithms, model export and multiple language interfaces like R and Python.

1 of 71
Download to read offline
H2O at Berlin R Meetup Introduction to ML with H2O and Deep Water Jo-fai (Joe) Chow Data Scientist joe@h2o.ai @matlabulous Berlin R at Adjust.com 23rd May, 2017
H2O at Berlin R Meetup Introduction to ML with H2O and Deep Water Jo-fai (Joe) Chow Data Scientist joe@h2o.ai @matlabulous Berlin R at Adjust.com 23rd May, 2017 All slides, data and code examples http://bit.ly/h2o_meetups
Agenda Introduction• Company• Why H• 2O? • H2O Machine Learning Platform Deep Water• Motivation / Benefits• GPU Deep Learning Demo• Sparkling Water• Kuba• ’s Talk Sparkling Water = H• 2O + Spark 3
About Me Civil (Water) Engineer• 2010• – 2015 Consultant (UK)• Utilities• Asset Management• Constrained Optimization• EngD (Industrial PhD) (UK)• Infrastructure Design Optimization• Machine Learning +• Water Engineering Discovered H• 2O in 2014 Data Scientist• 2015• – 2016 Virgin Media (UK)• Domino Data Lab (Silicon Valley)• 2016• – Present • H2O.ai (Silicon Valley) How?• bit.ly/joe_kaggle_story• 4
About Me – I ❤️ R 5
About Me – I ❤️ Colours 6 Developing R Packages for Fun rPlotter (2014)
About Me – I ❤️ DataViz 7 My First Data Viz & Shiny App Experience CrimeMap (2013) Revolution Analytics’ Data Viz Contest RUGSMAPS (2014)
useR! 2014 8 John Chambers mentioned H2O … so I gave it a try CrimeMap -> poster
About Me – I ❤️ Kaggle 9 R + H2O + Domino for Kaggle Guest Blog Post for Domino & H2O (2014) • The Long Story • bit.ly/joe_kaggle_story
About H2O.ai 10
Company Overview Founded 2011 Venture-backed, debuted in 2012 Products • H2O Open Source In-Memory AI Prediction Engine • Sparkling Water • Deep Water • Steam Mission Operationalize Data Science, and provide a platform for users to build beautiful data products Team 70 employees • Distributed Systems Engineers doing Machine Learning • World-class visualization designers Headquarters Mountain View, CA 11
12Joe (London)Kuba (Prague)
Scientific Advisory Council 13
14 Arno (CTO)
H2O Community Growth 15
16 http://www.h2o.ai/gartner-magic-quadrant/ Platforms with H2O Integrations
17 Check out our website h2o.ai
18 #AroundTheWorldWithH2Oai
19 Selfie Time Hello Berlin!
Why H2O? 20
Szilard Pafka’s ML Benchmark 21 https://github.com/szilard/benchm-ml n = million of samples Gradient Boosting Machine Benchmark H2O is fastest at 10M samples H2O is as accurate as others at 10M samples Time (s) AUC
H2O for Kaggle Competitions 22
H2O for Academic Research 23 http://www.sciencedirect.com/science/article/pii/S0377221716308657 https://arxiv.org/abs/1509.01199
H2O Machine Learning Platform 24
H2O.ai Offers AI Open Source Platform Product Suite to Operationalize Data Science with Visual Intelligence In-Memory, Distributed Machine Learning Algorithms with Speed and Accuracy H2O Integration with Spark. Best Machine Learning on Spark. 100% Open Source 25 Visual Intelligence and UX Framework For Data Interpretation and Story Telling on top of Beautiful Data Products Operationalize and Streamline Model Building, Training and Deployment Automatically and Elastically State-of-the-art Deep Learning on GPUs with TensorFlow, MXNet or Caffe with the ease of use of H2O Deep Water This Meetup
HDFS S3 NFS Distributed In-Memory Load Data Loss-less Compression H2O Compute Engine Production Scoring Environment Exploratory & Descriptive Analysis Feature Engineering & Selection Supervised & Unsupervised Modeling Model Evaluation & Selection Predict Data & Model Storage Model Export: Plain Old Java Object Your Imagination Data Prep Export: Plain Old Java Object Local SQL High Level Architecture 26
HDFS S3 NFS Distributed In-Memory Load Data Loss-less Compression H2O Compute Engine Production Scoring Environment Exploratory & Descriptive Analysis Feature Engineering & Selection Supervised & Unsupervised Modeling Model Evaluation & Selection Predict Data & Model Storage Model Export: Plain Old Java Object Your Imagination Data Prep Export: Plain Old Java Object Local SQL High Level Architecture 27 Import Data from Multiple Sources
HDFS S3 NFS Distributed In-Memory Load Data Loss-less Compression H2O Compute Engine Production Scoring Environment Exploratory & Descriptive Analysis Feature Engineering & Selection Supervised & Unsupervised Modeling Model Evaluation & Selection Predict Data & Model Storage Model Export: Plain Old Java Object Your Imagination Data Prep Export: Plain Old Java Object Local SQL High Level Architecture 28 Fast, Scalable & Distributed Compute Engine Written in Java
Supervised Learning Generalized Linear Models• : Binomial, Gaussian, Gamma, Poisson and Tweedie Naïve Bayes• Statistical Analysis Ensembles Distributed Random Forest• : Classification or regression models Gradient Boosting Machine• : Produces an ensemble of decision trees with increasing refined approximations Deep Neural Networks Deep learning• : Create multi-layer feed forward neural networks starting with an input layer followed by multiple layers of nonlinear transformations Algorithms Overview Unsupervised Learning • K-means: Partitions observations into k clusters/groups of the same spatial size. Automatically detect optimal k Clustering Dimensionality Reduction Principal Component Analysis• : Linearly transforms correlated variables to independent components Generalized Low Rank Models:• extend the idea of PCA to handle arbitrary data consisting of numerical, Boolean, categorical, and missing data Anomaly Detection Autoencoders• : Find outliers using a nonlinear dimensionality reduction using deep learning 29
Distributed Algorithms Foundation for In• -Memory Distributed Algorithm Calculation - Distributed Data Frames and columnar compression All algorithms are distributed in H• 2O: GBM, GLM, DRF, Deep Learning and more. Fine-grained map-reduce iterations. Only enterprise• -grade, open-source distributed algorithms in the market User Benefits Advantageous Foundation • “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 Parallel Parse into Distributed Rows Fine Grain Map Reduce Illustration: Scalable Distributed Histogram Calculation for GBM FoundationforDistributedAlgorithms 30
H2O Deep Learning in Action 31
HDFS S3 NFS Distributed In-Memory Load Data Loss-less Compression H2O Compute Engine Production Scoring Environment Exploratory & Descriptive Analysis Feature Engineering & Selection Supervised & Unsupervised Modeling Model Evaluation & Selection Predict Data & Model Storage Model Export: Plain Old Java Object Your Imagination Data Prep Export: Plain Old Java Object Local SQL High Level Architecture 32 Multiple Interfaces
H2O + R 33
Use H2O with Other R Packages 34 https://github.com/woobe/h2o_demos/tree/master/human_activitiy_recognition_with_smartphones Train a PCA model with H1. 2O Visualize PCs with plotly2. Deploy as a Shiny app3.
35 H2O Flow (Web) Interface
H2O + Python 36
HDFS S3 NFS Distributed In-Memory Load Data Loss-less Compression H2O Compute Engine Production Scoring Environment Exploratory & Descriptive Analysis Feature Engineering & Selection Supervised & Unsupervised Modeling Model Evaluation & Selection Predict Data & Model Storage Model Export: Plain Old Java Object Your Imagination Data Prep Export: Plain Old Java Object Local SQL High Level Architecture 37 Export Standalone Models for Production
38 docs.h2o.ai
H2O Tutorials Introduction to Machine• Learning with H2O Basic Extract, Transform and Load• (ETL) Supervised Learning• Parameters Tuning• Model Stacking• GitHub Repository• bit.ly/joe_h• 2o_tutorials Both R & Python Code Examples• Included 39
40
TensorFlow • Open source machine learning framework by Google • Python / C++ API • TensorBoard • Data Flow Graph Visualization • Multi CPU / GPU • v0.8+ distributed machines support • Multi devices support • desktop, server and Android devices • Image, audio and NLP applications • HUGE Community • Support for Spark, Windows … 41 https://github.com/tensorflow/tensorflow
42 https://github.com/dmlc/mxnet https://www.zeolearn.com/magazine/amazon-to-use-mxnet-as-deep-learning-framework
Caffe • Convolution Architecture For Feature Extraction (CAFFE) Pure C++ / CUDA architecture• for deep learning Command line, Python and• MATLAB interface Model Zoo• Open collection of models• 43 https://docs.google.com/presentation/d/1UeKXVgRvvxg9OUdh_UiC5G71UMscNPlvArsWER41PsU/
H2O Deep Learning in Action 44
Both TensorFlow and H2O are widely used 45
TensorFlow , MXNet, Caffe and H2O DL democratize the power of deep learning. H2O platform democratizes artificial intelligence & big data science. There are other open source deep learning libraries like Theano and Torch too. Let’s have a party, this will be fun! 46
Deep Water Next-Gen Distributed Deep Learning with H2O H2O integrates with existing GPU backends for significant performance gains One Interface - GPU Enabled - Significant Performance Gains Inherits All H2O Properties in Scalability, Ease of Use and Deployment Recurrent Neural Networks enabling natural language processing, sequences, time series, and more Convolutional Neural Networks enabling Image, video, speech recognition Hybrid Neural Network Architectures enabling speech to text translation, image captioning, scene parsing and more Deep Water 47
48 www.h2o.ai/download/
Available Networks in Deep Water LeNet• AlexNet• VGGNet• Inception (GoogLeNet)• ResNet (Deep Residual• Learning) Build Your Own• 49 ResNet
Unified Interface (Deep Water + R) 50 Choosing different network structures and backends
51 Example: Deep Water + H2O Flow Choosing different network structures
52 Choosing different backends (TensorFlow, MXNet, Caffe)
Unified Interface (Deep Water + Python) 53 Change backend to “mxnet”, “caffe” or “auto” Choosing different network structures
54 https://github.com/h2oai/h2o-3/tree/master/examples/deeplearning/notebooks Deep Water Example notebooks
Deep Water Cat/Dog/Mouse Demo 55
Deep Water R Demo • H2O + MXNet + TensorFlow • Dataset – Cat/Dog/Mouse • MXNet & TensorFlow as GPU backend • Train LeNet (CNN) models • R Demo Code and Data• github.com/h• 2oai/deepwater 56
Data – Cat/Dog/Mouse Images 57
Data – CSV 58
Deep Water – Basic Usage Live Demo if Possible 59
Import CSV 60
Train a CNN (LeNet) Model on GPU 61 Easy Switch
Train a CNN (LeNet) Model on GPU 62 Using GPU for training
Model 63
Deep Water – Custom Network 64
Xxx 65 Saving the custom network structure as a file Configure custom network structure (MXNet syntax)
Train a Custom Network 66 Point it to the custom network structure file
Model 67 Note: Overfitting is expected as we only use a very small datasets to demonstrate the APIs only
Conclusions 68
Project “Deep Water” • H2O + TF + MXNet + Caffe A powerful combination of widely• used open source machine learning libraries. All Goodies from H• 2O Inherits all H• 2O properties in scalability, ease of use and deployment. Unified Interface• Allows users to build, stack and• deploy deep learning models from different libraries efficiently. 69 Latest Nightly Build• https://s• 3.amazonaws.com/h2o- deepwater/public/nightly/latest/h 2o.jar 100• % Open Source The party will get bigger!•
Other H2O Developments • H2O + xgboost [Link] • Stacked Ensembles [Link] • Automatic Machine Learning [Link] • Time Series [Link] High Availability Mode in• Sparkling Water [Link] word• 2vec [Link] 70
Organizers & Sponsors• Berry Boessenkool & Berlin R• Nikola Chochkov & Adjust.com• 71 Danke! Code, Slides & Documents• bit.ly/h• 2o_meetups docs.h• 2o.ai Contact• joe@h• 2o.ai @matlabulous• github.com/woobe• Please search/ask questions on• Stack Overflow Use the tag `• h2o` (not H2 zero)

More Related Content

Berlin R Meetup

  • 1. H2O at Berlin R Meetup Introduction to ML with H2O and Deep Water Jo-fai (Joe) Chow Data Scientist joe@h2o.ai @matlabulous Berlin R at Adjust.com 23rd May, 2017
  • 2. H2O at Berlin R Meetup Introduction to ML with H2O and Deep Water Jo-fai (Joe) Chow Data Scientist joe@h2o.ai @matlabulous Berlin R at Adjust.com 23rd May, 2017 All slides, data and code examples http://bit.ly/h2o_meetups
  • 3. Agenda Introduction• Company• Why H• 2O? • H2O Machine Learning Platform Deep Water• Motivation / Benefits• GPU Deep Learning Demo• Sparkling Water• Kuba• ’s Talk Sparkling Water = H• 2O + Spark 3
  • 4. About Me Civil (Water) Engineer• 2010• – 2015 Consultant (UK)• Utilities• Asset Management• Constrained Optimization• EngD (Industrial PhD) (UK)• Infrastructure Design Optimization• Machine Learning +• Water Engineering Discovered H• 2O in 2014 Data Scientist• 2015• – 2016 Virgin Media (UK)• Domino Data Lab (Silicon Valley)• 2016• – Present • H2O.ai (Silicon Valley) How?• bit.ly/joe_kaggle_story• 4
  • 5. About Me – I ❤️ R 5
  • 6. About Me – I ❤️ Colours 6 Developing R Packages for Fun rPlotter (2014)
  • 7. About Me – I ❤️ DataViz 7 My First Data Viz & Shiny App Experience CrimeMap (2013) Revolution Analytics’ Data Viz Contest RUGSMAPS (2014)
  • 8. useR! 2014 8 John Chambers mentioned H2O … so I gave it a try CrimeMap -> poster
  • 9. About Me – I ❤️ Kaggle 9 R + H2O + Domino for Kaggle Guest Blog Post for Domino & H2O (2014) • The Long Story • bit.ly/joe_kaggle_story
  • 11. Company Overview Founded 2011 Venture-backed, debuted in 2012 Products • H2O Open Source In-Memory AI Prediction Engine • Sparkling Water • Deep Water • Steam Mission Operationalize Data Science, and provide a platform for users to build beautiful data products Team 70 employees • Distributed Systems Engineers doing Machine Learning • World-class visualization designers Headquarters Mountain View, CA 11
  • 21. Szilard Pafka’s ML Benchmark 21 https://github.com/szilard/benchm-ml n = million of samples Gradient Boosting Machine Benchmark H2O is fastest at 10M samples H2O is as accurate as others at 10M samples Time (s) AUC
  • 22. H2O for Kaggle Competitions 22
  • 23. H2O for Academic Research 23 http://www.sciencedirect.com/science/article/pii/S0377221716308657 https://arxiv.org/abs/1509.01199
  • 24. H2O Machine Learning Platform 24
  • 25. H2O.ai Offers AI Open Source Platform Product Suite to Operationalize Data Science with Visual Intelligence In-Memory, Distributed Machine Learning Algorithms with Speed and Accuracy H2O Integration with Spark. Best Machine Learning on Spark. 100% Open Source 25 Visual Intelligence and UX Framework For Data Interpretation and Story Telling on top of Beautiful Data Products Operationalize and Streamline Model Building, Training and Deployment Automatically and Elastically State-of-the-art Deep Learning on GPUs with TensorFlow, MXNet or Caffe with the ease of use of H2O Deep Water This Meetup
  • 26. HDFS S3 NFS Distributed In-Memory Load Data Loss-less Compression H2O Compute Engine Production Scoring Environment Exploratory & Descriptive Analysis Feature Engineering & Selection Supervised & Unsupervised Modeling Model Evaluation & Selection Predict Data & Model Storage Model Export: Plain Old Java Object Your Imagination Data Prep Export: Plain Old Java Object Local SQL High Level Architecture 26
  • 27. HDFS S3 NFS Distributed In-Memory Load Data Loss-less Compression H2O Compute Engine Production Scoring Environment Exploratory & Descriptive Analysis Feature Engineering & Selection Supervised & Unsupervised Modeling Model Evaluation & Selection Predict Data & Model Storage Model Export: Plain Old Java Object Your Imagination Data Prep Export: Plain Old Java Object Local SQL High Level Architecture 27 Import Data from Multiple Sources
  • 28. HDFS S3 NFS Distributed In-Memory Load Data Loss-less Compression H2O Compute Engine Production Scoring Environment Exploratory & Descriptive Analysis Feature Engineering & Selection Supervised & Unsupervised Modeling Model Evaluation & Selection Predict Data & Model Storage Model Export: Plain Old Java Object Your Imagination Data Prep Export: Plain Old Java Object Local SQL High Level Architecture 28 Fast, Scalable & Distributed Compute Engine Written in Java
  • 29. Supervised Learning Generalized Linear Models• : Binomial, Gaussian, Gamma, Poisson and Tweedie Naïve Bayes• Statistical Analysis Ensembles Distributed Random Forest• : Classification or regression models Gradient Boosting Machine• : Produces an ensemble of decision trees with increasing refined approximations Deep Neural Networks Deep learning• : Create multi-layer feed forward neural networks starting with an input layer followed by multiple layers of nonlinear transformations Algorithms Overview Unsupervised Learning • K-means: Partitions observations into k clusters/groups of the same spatial size. Automatically detect optimal k Clustering Dimensionality Reduction Principal Component Analysis• : Linearly transforms correlated variables to independent components Generalized Low Rank Models:• extend the idea of PCA to handle arbitrary data consisting of numerical, Boolean, categorical, and missing data Anomaly Detection Autoencoders• : Find outliers using a nonlinear dimensionality reduction using deep learning 29
  • 30. Distributed Algorithms Foundation for In• -Memory Distributed Algorithm Calculation - Distributed Data Frames and columnar compression All algorithms are distributed in H• 2O: GBM, GLM, DRF, Deep Learning and more. Fine-grained map-reduce iterations. Only enterprise• -grade, open-source distributed algorithms in the market User Benefits Advantageous Foundation • “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 Parallel Parse into Distributed Rows Fine Grain Map Reduce Illustration: Scalable Distributed Histogram Calculation for GBM FoundationforDistributedAlgorithms 30
  • 31. H2O Deep Learning in Action 31
  • 32. HDFS S3 NFS Distributed In-Memory Load Data Loss-less Compression H2O Compute Engine Production Scoring Environment Exploratory & Descriptive Analysis Feature Engineering & Selection Supervised & Unsupervised Modeling Model Evaluation & Selection Predict Data & Model Storage Model Export: Plain Old Java Object Your Imagination Data Prep Export: Plain Old Java Object Local SQL High Level Architecture 32 Multiple Interfaces
  • 34. Use H2O with Other R Packages 34 https://github.com/woobe/h2o_demos/tree/master/human_activitiy_recognition_with_smartphones Train a PCA model with H1. 2O Visualize PCs with plotly2. Deploy as a Shiny app3.
  • 35. 35 H2O Flow (Web) Interface
  • 37. HDFS S3 NFS Distributed In-Memory Load Data Loss-less Compression H2O Compute Engine Production Scoring Environment Exploratory & Descriptive Analysis Feature Engineering & Selection Supervised & Unsupervised Modeling Model Evaluation & Selection Predict Data & Model Storage Model Export: Plain Old Java Object Your Imagination Data Prep Export: Plain Old Java Object Local SQL High Level Architecture 37 Export Standalone Models for Production
  • 39. H2O Tutorials Introduction to Machine• Learning with H2O Basic Extract, Transform and Load• (ETL) Supervised Learning• Parameters Tuning• Model Stacking• GitHub Repository• bit.ly/joe_h• 2o_tutorials Both R & Python Code Examples• Included 39
  • 40. 40
  • 41. TensorFlow • Open source machine learning framework by Google • Python / C++ API • TensorBoard • Data Flow Graph Visualization • Multi CPU / GPU • v0.8+ distributed machines support • Multi devices support • desktop, server and Android devices • Image, audio and NLP applications • HUGE Community • Support for Spark, Windows … 41 https://github.com/tensorflow/tensorflow
  • 43. Caffe • Convolution Architecture For Feature Extraction (CAFFE) Pure C++ / CUDA architecture• for deep learning Command line, Python and• MATLAB interface Model Zoo• Open collection of models• 43 https://docs.google.com/presentation/d/1UeKXVgRvvxg9OUdh_UiC5G71UMscNPlvArsWER41PsU/
  • 44. H2O Deep Learning in Action 44
  • 45. Both TensorFlow and H2O are widely used 45
  • 46. TensorFlow , MXNet, Caffe and H2O DL democratize the power of deep learning. H2O platform democratizes artificial intelligence & big data science. There are other open source deep learning libraries like Theano and Torch too. Let’s have a party, this will be fun! 46
  • 47. Deep Water Next-Gen Distributed Deep Learning with H2O H2O integrates with existing GPU backends for significant performance gains One Interface - GPU Enabled - Significant Performance Gains Inherits All H2O Properties in Scalability, Ease of Use and Deployment Recurrent Neural Networks enabling natural language processing, sequences, time series, and more Convolutional Neural Networks enabling Image, video, speech recognition Hybrid Neural Network Architectures enabling speech to text translation, image captioning, scene parsing and more Deep Water 47
  • 49. Available Networks in Deep Water LeNet• AlexNet• VGGNet• Inception (GoogLeNet)• ResNet (Deep Residual• Learning) Build Your Own• 49 ResNet
  • 50. Unified Interface (Deep Water + R) 50 Choosing different network structures and backends
  • 51. 51 Example: Deep Water + H2O Flow Choosing different network structures
  • 53. Unified Interface (Deep Water + Python) 53 Change backend to “mxnet”, “caffe” or “auto” Choosing different network structures
  • 56. Deep Water R Demo • H2O + MXNet + TensorFlow • Dataset – Cat/Dog/Mouse • MXNet & TensorFlow as GPU backend • Train LeNet (CNN) models • R Demo Code and Data• github.com/h• 2oai/deepwater 56
  • 59. Deep Water – Basic Usage Live Demo if Possible 59
  • 61. Train a CNN (LeNet) Model on GPU 61 Easy Switch
  • 62. Train a CNN (LeNet) Model on GPU 62 Using GPU for training
  • 64. Deep Water – Custom Network 64
  • 65. Xxx 65 Saving the custom network structure as a file Configure custom network structure (MXNet syntax)
  • 66. Train a Custom Network 66 Point it to the custom network structure file
  • 67. Model 67 Note: Overfitting is expected as we only use a very small datasets to demonstrate the APIs only
  • 69. Project “Deep Water” • H2O + TF + MXNet + Caffe A powerful combination of widely• used open source machine learning libraries. All Goodies from H• 2O Inherits all H• 2O properties in scalability, ease of use and deployment. Unified Interface• Allows users to build, stack and• deploy deep learning models from different libraries efficiently. 69 Latest Nightly Build• https://s• 3.amazonaws.com/h2o- deepwater/public/nightly/latest/h 2o.jar 100• % Open Source The party will get bigger!•
  • 70. Other H2O Developments • H2O + xgboost [Link] • Stacked Ensembles [Link] • Automatic Machine Learning [Link] • Time Series [Link] High Availability Mode in• Sparkling Water [Link] word• 2vec [Link] 70
  • 71. Organizers & Sponsors• Berry Boessenkool & Berlin R• Nikola Chochkov & Adjust.com• 71 Danke! Code, Slides & Documents• bit.ly/h• 2o_meetups docs.h• 2o.ai Contact• joe@h• 2o.ai @matlabulous• github.com/woobe• Please search/ask questions on• Stack Overflow Use the tag `• h2o` (not H2 zero)