Data Science in Enterprise

1© Cloudera, Inc. All rights reserved.
Josh YEH | Software Engineer
https://www.linkedin.com/in/joshyeh/
Data Science in the Enterprise

This is the age of machine learning.
2
Cost of compute
Data volume
Time
Machine
Learning
NO
Machine
Learning
1950
s
1960
s
1970
s
1980
s
1990
s
2000
s
2010
s

CDSW
Data Scientist
Machine Learning and
Advanced Analytical
Platform
CM / CDH
Data Engineer
Data Admin
ETL and ...

Machine learning opportunities are everywhere.
Data Engineering Data Science (Exploratory) Production (Operational)
Data Wrangling
Visualization
and Analysis
Model Training
& Testing
Production
Data Pipelines Batch Scoring
Online Scoring
Serving
Data GovernanceGovernance
Processing
Acquisition
Reports,
Dashboards
Data has never been
more plentiful
Open source data science and
machine learning libraries are
rapidly evolving
Commodity (and on-demand) compute
makes scalable production machine
learning affordable

But, there are challenges:
Data Wrangling
Visualization
and Analysis
Model Training
& Testing
Production
Online Scoring
Serving
Processing
Acquisition
Reports,
Dashboards
Most data science done at
small scale, individually,
and is difficult to replicate
Very few models
reach production
Teams have different,
conflicting requests for
languages & libraries
Data needs to move
across multiple different
systems

Our enterprise customers say:
Access
For sensitive data, secure clusters are
difficult to access. And IT typically
doesn’t want random packages
installed on a secure cluster.
Popular open source tools don’t easily
connect to these environments, or
always support Hadoop data formats.
Scale
Laptops rarely have capacity for
medium, let alone big data. This
leads to a lot of sampling.
Popular frameworks don’t easily
parallelize on a cluster. Typically
code has to get rewritten for
production.
Developer Experience
Notebooks, while awesome, don’t
easily support virtual environment
and dependency management,
especially for teams. This makes
sharing and reproducibility hard.
Notebooks are also challenging to
“put into production.”

Cloudera: The Enterprise Platform for
Data Science and Machine Learning
The data is now here
30B
CONNECTED DEVICES
440x
MORE DATA
Cloudera first to integrate Spark
Modern Platform for Machine Learning and Advanced Analytics
Leading adoption among enterprises
500Customers
Run Spark on

Our goal: an open data science at enterprise scale
Help more data scientists
use the power of Cloudera
Use a powerful, familiar
environment with direct access to
Cloudera data and compute
Data Scientist
Data Engineer
Make it easy and secure to
add new users, use cases
Offer secure self-service analytics
and a faster path to production on
common, affordable infrastructure
Enterprise Architect
Hadoop Admin

An open ecosystem for agility and innovation
Open Ecosystem Black Box

Rich data science libraries and the power of Spark
IT
drive adoption while maintaining compliance
Data Scientist
explore, experiment, iterate

Supports the complete data science pipeline
From data to exploration to action
Data Wrangling
Visualization
and Analysis
Model Training
& Testing
Production
Online Scoring
Serving
Processing
Acquisition
Reports,
Dashboards
CDSW
Data Scientist
Machine Learning and
Advanced Analytical Platform
CM / CDH
Data Engineer
Data Administrator,
ETL and ...

Data science at enterprise scale requires a full stack.
• Support unlimited data
• Provide sufficient tools for Analysts
• Provide sufficient tools for
Data Scientists + Data Engineers
• Enable real-time use cases
• Provide data governance
• Provide full-stack security
• Deploy in the cloud
• Integrate with partner tools
• Be easy for IT to deploy/maintain
✓Hadoop
✓Impala, Hive, Hue
✓Spark, Data Science Workbench
✓Kafka, Spark Streaming
✓Navigator + Partners
✓Kerberos, Sentry, Record Service, KMS/KTS
✓Cloudera Director
✓Rich Ecosystem
✓Cloudera Manager + Director

Introducing Cloudera Data Science Workbench
Self-service data science for the enterprise
Accelerates data science from
development to production with:
• Secure self-service environments
for data scientists to work against
Cloudera clusters
• Support for Python, R, and Scala,
plus project dependency isolation
for multiple library versions
• Workflow automation, version
control, collaboration and sharing

Data scientists can:
• Use R, Python, or Scala from a web
browser, with no desktop footprint
• Install any library or framework within
isolated project environments
• Directly access data in secure clusters
with Spark and Impala
• Share insights with their team for
reproducible, collaborative research
• Automate and monitor data pipelines
using built-in job scheduling
IT can:
• Give their data science team the
freedom to work how they want, when
they want
• Stay compliant with out-of-the-box
support for full platform security,
especially Kerberos
• Run on-premises or in the cloud,
wherever data is managed
With Cloudera Data Science Workbench…

Enable Data Scientists IT can focus data engineering
With Cloudera Data Science Workbench…

Demo
CDH security integration, Share Report, Group Collaboration
Data Visualization, Abnormality Detection,
GPU resource management, Train ML models

Connect to Kerberized Hadoop cluster with
• As easy as one single
configuration for a given
user.
• CDSW application will
refresh kerberos
authentication every 15
mins
• Support both Kerberos
password and key tab
authentication.
• Revoke is easy.

Cloudera Fast Forward Lab (CFFL)
Machine Learning Kickstart kit

What is Cloudera Fast Forward Labs (CFFL)
Fast Forward Labs (FFL) is a small data science/machine learning research and
consulting team.
FFL provides a subscription advisory service that helps technology leaders keep
current on machine learning and artificial intelligence innovations that will be
practical within the next 6-24 months

1.Executives struggle with machine learning strategy
There are thousands of different ML
techniques available, and innovation is
incredibly fast. It's impossible for any
company know about all the options, let alone
evaluate the few that may meet their needs.
The vendor ecosystem is increasingly complex,
and hype muddies the waters. Analyst firms
like Gartner and Forrester take a high-level
and long view of the market, so they don't
provide any real actionable intelligence about
ML that could be useful today.

2. Cut through the ML hype
A Cloudera Fast Forward Labs research subscription is the
best way to keep up with the latest in machine learning and
artificial intelligence. Each quarter, we publish a research
report and provide software prototypes highlighting
emerging machine learning capabilities that will be useful in
the next two years. The prototypes prove the algorithms,
reduce risk and uncertainty, and shorten time to value. Our
advising services expand on the research and help you apply
machine learning to your specific line of work, keeping you
ahead of the curve and above the confusion and hype.

3. Research and so much more

CFFL Customer Successes
• A bank's head of innovation read the CFFL report on Natural Language
Generation. He is now generating 80% of its regulatory compliance
documentation automatically, saving 30% of the hours.
• One of the big four accounting firms is using Summarization techniques to
reissue tax guidance to its customers. What were once static memos, are now
dynamically updated documents. Their ML system sits on a newsfeed of
updates to judicial interpretation of the tax law, automatically alerting the CPA
and their customers.
• Another customer read our report on Image and video analysis with deep
learning. CFFL conducted a feasibility study to get surgical robots to be able
detect when a surgery is going wrong. The findings wound up radically
changing the company's product roadmap.

Future Data Science Trending
Invest $$ on the right direction; avoid building technical debts

Trending 1: GPU for training; CPU for deployment
1. GPU is built for Machine Learning ( Deep
Neural Network Learning / AI ).
2. Data retention (/revamp) is costly.
a. Eg: the data from the development,
test, manufacture process for chips
that are used in autonomous car, need
to need 15 years of retention.
3. Re-train / Re-deploy new DL model with
more and more data set daily, weekly,
yearly etc...

Trending 2: Investment on Right Tool Sets
1. Don’t reinvent the wheel: Leverage
what’s available out there to shorten
development cycle.
2. Use community support: Mainstream
tools have ample enthusiastic support and
contribution.
3. Easier troubleshooting

Trending 3: Application Containerization
1. Maintenance: legacy application,
complicated framework, 3rd party
libraries, evaluate free open projects. Too
many things could go wrong.
2. Resource Optimization: easy resource
management via cgroup in cluster.
3. Deployment: Easy testing, validation and
deployment.
4. Scale: kubernete to manage containers
and easy scale-up.

Trending 4: Simplified Data Scientist Workflow
1. Work Anywhere 2. Train at large scale 3. Seamless deployment
90+% resource spent on

Thank you!
jjyeh@cloudera.com

Data Science in Enterprise

More Related Content

What's hot (20)

Similar to Data Science in Enterprise (20)

Recently uploaded (20)

Data Science in Enterprise