Vous êtes sur la page 1sur 34

Google Cloud Platform

Cloud Machine Learning


Marc Anthony Reyes (@marcreyesph)
Data is exploding.
And smart companies are taking advantage.
Unstructured data accounts for
90% of enterprise data*
Cloud Machine Learning helpyou
make sense of it

*Source: IDC
What is Machine Learning?

Data Algorithm Insight

Google CloudPlatform Confidential &Proprietary 4


Machine Learning @Google

Google CloudPlatform Confidential &Proprietary 5


Pool
Google Translate

Google CloudPlatform Confidential &Proprietary 7


Google CloudPlatform Confidential &Proprietary 8
The point of ML is to make predictions

Input Feature Predicted Value

Model

Google CloudPlatform Confidential &Proprietary 9


Tensorflow helps you “train”models

Input Feature Predicted Value

Model

Cost

Update model
based on Cost True Value

Google CloudPlatform Confidential &Proprietary 10


Democratizing machine learning

CloudML ML APIs
Translate API

Vision API

Speech API
Scale,No-ops
Infrastructure Language API

Use/extend OSSSDK Build custom models Use pre-built models

ML researcher Data Scientist App Developer

Google CloudPlatform Confidential &Proprietary 11


Beyond Tensorflow

Scale of CloudML (a) ML APIs


Compute
Problem Deep networks Translate API

TensorFlow Vision API


Accuracy Processing
Units (TPUs) Speech API

Distributed
Language API
No-ops

Size ofdataset
Size ofNN
https://cloudplatform.googleblog.com/2016/05/Google-supercharges-machine-learning-tasks-with-custom-
chip.html
Google CloudPlatform Confidential &Proprietary 12
ML APIs are simply REST callsand can be made
from any language orframework

sservice = build('speech', 'v1beta1', developerKey=APIKEY)


response = sservice.speech().syncrecognize(
body={
'config': {
'encoding': 'LINEAR16',
'sampleRate': 16000
},
'audio': {
'uri': 'gs://cloud-training-demos/vision/audio.raw' Data on CloudStorage
}
}).execute()
print response

Google CloudPlatform Confidential &Proprietary 13


Enterprise Predictive Analytics Challenges

Data access to avariety Data preparation, Develop and build


of data sources. exploration and visualization. analytic models.

Deploy models and integrate Perform platform, project High performance and scalability
them into businessprocesses and model management. for both development
and applications. and deployment.

Google CloudPlatform Confidential &Proprietary 14


Data Warehouse is theFoundation
of Something Bigger

Data Machine Intelligence Predictive


Warehouses/Lakes +
Prescriptive
Train
Cloud
YourOwn Analytics
Models
=
Advanced
Machine Analytics
On Learning
Premises APIs

Google CloudPlatform Confidential &Proprietary 15


Machine Learning UseCases
Manufacturing Retail Healthcare and Life Sciences

•Predictive maintenance or condition • Predictive inventory planning •Alerts and diagnostics from real-time
monitoring • Recommendation engines patient data
•Warranty reserve estimation • Upsell and cross-channel marketing •Disease identification and risk satisfaction
•Propensity to buy • Market segmentation andtargeting •Patient triage optimization
•Demand forecasting • Customer ROI and lifetime value •Proactive health management
•Process optimization •Healthcare provider sentiment analysis

Travel and Hospitality Financial Services Energy, Feedstock and Utilities

•Aircraft scheduling •Risk analytics and regulation • Power usage analytics


•Dynamic pricing •Customer Segmentation • Seismic data processing
•Social media – consumer feedback and •Cross-selling and up-selling • Carbon emissions and trading
interaction analysis •Sales and marketing campaign • Customer-specific pricing
•Customer complaint resolution management • Smart grid management
•Traffic patterns andcongestion •Credit worthiness evaluation • Energy demand and supply optimization
management

Google CloudPlatform Confidential &Proprietary 16


Why So Little Machine Learning Apps OutThere?

• Building and scaling machine learning infrastructureis


hard

• Operating production ML system is time consuming and


expensive

Google CloudPlatform Confidential &Proprietary 17


Building Smart Applications Today

Technology Operationalization Tooling

Difficult to scale Complex data pipelines Complex dev pipelinewith


many combinations of
Many choices for different Managing ML infra takes tools/libraries
use cases away time from actually
doing ML Not fully interactive
Using latest technology (e.g. developer experience -
DNN) is hard Many models to manage collaboration/sharing is hard

Google CloudPlatform Confidential &Proprietary 18


Introducing Cloud Machine Learning
● Fully managed service
● Train using a custom TensorFlowgraph
for any ML usecases
● Training at scale to shorten dev cycle
● Automatically maximize predictive
accuracy with HyperTune
● Batch and online predictions, atscale
● Integrated Datalab experience

Google CloudPlatform Confidential &Proprietary 19


Powerful Machine Learning Algorithm

● Convolutional Neural Network for image classification


● Recursive Neural network fortext sentiment analysis
● Linear regression at scale to predict consumer action
(purchase prediction, churn analysis)
● And unlimited variety of algorithms you can build using TensorFlow

Google CloudPlatform Confidential &Proprietary 21


Automatically tune your model withHyperTune

● Automatic hyperparameter tuning Want to findthis

service Not these


Build better performingmodels

Objective

faster and save many hours of
manual tuning
● Google-developed search
algorithm efficiently finds
better hyperparameters for
your model/dataset

Google CloudPlatform Confidential &Proprietary 22


Integrated with GCPProducts

● Access data that is stored in GCS or BigQuery


● Save trained models to GCS
● Preprocess largest datasets (TB) usingDataflow
● Orchestrate ML workflowas a Dataflow pipeline
● Analyze data and interactively develop ML models in Datalab
● AutoML for premade and customized ML models curated for
your specific ML projects

Google CloudPlatform Confidential &Proprietary 23


Fully Managed Machine Learning Services

● Scalable and distributed training infrastructure foryour largest


data sets
● Scalable prediction infrastructure thatcan serve very large traffic
● Managed no-ops infrastructure handles provisioning, scaling,
and monitoring so that you can focus on building your models
instead of handlingclusters

Google CloudPlatform Confidential &Proprietary 24


Transform Data into Actions
Data Advanced
Data Exploration&
Preparation& Databases Storage Analytics Analytics &
Ingestion Collaboration
Processing Intelligence

Data Relational
Mobileapps preparation
Development
Logs environment Business
Data Federated for Machine analysts
Key-value exploration query Learning

Data scientists
Webapps Stream Document Object
processing

Messaging
Data Pre-Trained Developers
visualization Data catalog
Widecolumn Machine
Learning
Sensorsand models
devices
Batch
SQL
processing

Google CloudPlatform Confidential &Proprietary 29


Transform Data into Actions
Data Advanced
Databases Exploration &
Data Ingestion Preparation& Analytics Analytics &
/ Storage Collaboration
Processing Intelligence

Mobileapps
Cloud SQL Cloud Datalab GoogleBigQuery Cloud Machine
Learning Business
analysts

App Engine Cloud Dataflow


Cloud Google Data scientists
Webapps CloudDataproc
Datastore BigQuery Speech API

Developers
Cloud Pub/Sub CloudDataproc CloudDataproc GoogleDrive
Cloud Bigtable Vision API
Sensorsand
devices

Google Google
CloudStorage Analytics 360 Analytics 360 TranslateAPI

Google CloudPlatform Confidential &Proprietary 30


Use Your Own Data to Train Models

BETA

Cloud MachineLearning

GA GA BETA

Cloud Storage Google BigQuery CloudDatalab Develop/Model/Test

Google CloudPlatform Confidential &Proprietary 31


Use your own data to train models
Training flow

DataStorage Pre-Processing Training Tooling Local


training

Datalab

Upload
Hosted Model
Prediction flow Download

Batch
Prediction Tooling
Mobile
predictio
n
HTTP request Datalab
Online

Google CloudPlatform Confidential &Proprietary 32


Machine Intelligence is Already Making a Huge Difference
and There are Many, Many MoreOpportunities

1 2 3 4
Identify categorizations Capture thousands of Evaluate the model by
Automatically
that provide value, examples of correct applying it against
categorize, and
categories you’re evaluations for that additional manually
automatically
already evaluating for categorization, and use categorized data, correct
extract value
by hand today them to train an ML model and tune

Google CloudPlatform Confidential &Proprietary 33


Introduction to GCP Cloud
ML with Qwiklabs

Google CloudPlatform Confidential &Proprietary 35


Codelabs this Afternoon

● Cloud ML Engine: Qwik Start


● Cloud Natural Language API: Qwik Start

Google CloudPlatform Confidential &Proprietary 36


Awesome Stuff for Participants

● One-month free access to Qwiklabs with 150 credits for free


platform usage.
● One-month free access to GCP Introductory Course on
Coursera.

Google CloudPlatform Confidential &Proprietary 37


Access These Links for the Qwiklab Activity

● Baseline: Data, ML, AI Quest


● Cloud ML Engine: Qwik Start (http://bit.ly/GCP18CDOCloudML)
● Cloud Natural Language API: Qwik Start
(http://bit.ly/GCP18CDONL)
● This afternoon, we only will be taking Cloud ML Engine: Qwik
Start and Cloud Natural Language API: Qwik Start

Google CloudPlatform Confidential &Proprietary 38


Let’s get our hands dirty with
code

Google CloudPlatform Confidential &Proprietary 39


cloud.google.com

Vous aimerez peut-être aussi