Project “Deep Water” (H2O integration with other deep learning libraries - Jo-Fai (Joe) Chow, H2O

Deep Water

Jo-fai (Joe) Chow

Data Scientist

[email protected]

@matlabulous

Data Science Milan

Politecnico di Milano

10th October, 2016

Agenda

• First Talk (25 mins)o About H2O.aio Demo

• A Simple Classification Task• H2O’s Web Interface

o Why H2O?• Our Community• Our Customers

o What’s Next?• New H2O Features

• Second Talk (25 mins)o H2O for IoT

• Predictive Maintenance• Anomaly Detection• H2O’s R Interface

• Third Talk (25 mins)o Deep Watero Demo

• H2O + mxnet on GPU• H2O’s Python Interface

2

Deep Learning in H2O

H2O Overview

4

A Simple Neural Network

5Image credit: http://cs231n.github.io/

H2O Deep Learning in Action

6

H2O Deep Learning Community Quotes

7

Why Deep Water?

D e e p Wa te r : N ex t - G e n D e e p L e a r n i n g i n H 2 O

9

D e e p W a t e r o p e n s t h e F l o o d g a t e s f o r S t a t e - o f - A r t D e e p L e a r n i n g

10

D e e p W a t e r o p e n s t h e F l o o d g a t e s f o r S t a t e - o f - A r t D e e p L e a r n i n g

11

Deep Water Demo

Deep Water Demo

• H2O + mxnet o Dataset:

• Cat / Dog / Mouse

o H2O Python interface

o mxnet GPU backend

o Train a LeNet (CNN) model

o Explore model in Flow

• Code and Datao bit.ly/h2o_milan_1

o subfolder• deep_water_demo

13

Data – Cat/Dog /Mouse Images

14

Data - CSV

15

H2O + mxnet Demo

17

bit.ly/h2o_milan_1Subfolder: deep_water_demo

18

“htop”(monitoring CPU/RAM)

“gpustat”(monitoring GPU)

“java –jar h2o.jar”H2O JVM Cluster

Deep Water Edition

“python demo_01_lenet.py”Python script for demo

19

H2O’s Python Module

Deep Water module

Connect to H2O Cluster

Import CSV

Define LeNet model in Deep Water

Train and show model

20

CPU for other tasks Using GPU for LeNetmodel training

H2O JVM Cluster

“python demo_01_lenet.py”Training LeNet Model

Note: H2O’s JVM on GPU

21

“python demo_01_lenet.py”Showing LeNet Model

22

Using Flow (localhost:54321) to explore data frame and model

23

Using Flow (localhost:54321) to explore data frame and model

24

Using Flow (localhost:54321) to split data and train Deep Water model

25

Using Flow to train Deep Water Model

26

Using Flow to train Deep Water Model

27

Choosing Different Network Structure

28

Choosing Different Backend

29

Training Deep Water Models without Programming

H2O’s Mission

30

Making Machine Learning Accessible to Everyone

Photo credit: Virgin Media

Grazie mil le!

31

• Data Science Milan

• Gianmario Spacagna

• Politecnico di Milano

• Resourceso bit.ly/h2o_milan_1

o www.h2o.ai

o docs.h2o.ai

• Contacto [email protected]

o @matlabulous

o github.com/woobe