3. How to..
From our connected
assets
generate new,
better services
for our customers
and increase our
business revenues ?
4. The gold is on
our data
But we need to extract it !
Source : Irish Business Network
5. There is a whole chain of complexity to manage and understand
A. Sauvage
Layouts
Data consumption
Data analytics, Machine Learning
Data presentation
Request engine performance
Storage capacity
Secure and resilient data flow pipe
Quality management
Data creation process (SCADA)
IoT project
integrated
architecture
Cloud.Config M2M
managed resources
On premise
management
6. What are the challenges in data science
development and deployment
• How do you … ?
• Share your data science project with your
team
• Ensure that you can reproduce your analysis
• Build a reliable and continuous data flow to
feed your model
• Manage the quality of your inputs
• Deploy and scale your project in production
environment
A. Sauvage
7. From local projects to Cloud managed models
A. Sauvage
DevOps
Azure
Cluster
Container 1
Container 2
Container 3
Project 3Project 2Project 1
Azure CloudData scientists
Data science development Data science development deployment
Project 3Project 2Project 1
Local development (Laptop)
ML workbench
8. Local build with Anaconda and Cloud deployment
A. Sauvage
Anaconda workbench
Python kernel + dep.
Project.ipynb
with ML
model
Jupyter Notebook
Local development (Laptop)
Azure
Cluster
Project 1 – container_02
Azure Cloud
web services
Project 1 – container_01
Git repo.
Project.ipynb
with ML model
Dockerfile
build structure
mounted volume
Jupyter/
kernel_gateway
image name
Docker image
Docker registry
Cloud services on top of Azure
Docker containers
Local development
9. Easy build and load with docker-compose
A. Sauvage
$>docker-compose up –d
$>docker ps
We can see here that 2 containers have been started
- The first one is for the Jupyter notebook itself so you can connect remotely to your Machine Learning
project and use the web browser to work on your model.
- The second one run a web server to host your Machine Learning model web service API
10. Access your Jupyter notebook
A. Sauvage
Note:
check the docker container
logs to find the token if you
haven’t setup any security
password
11. Machine Learning Development
- Notebook chapters -
A. Sauvage
• Chapter I : Dataset preparation and creation of the model features
• Chapter II : Create the multilinear regression model
• Chapter III : Evaluate the model performance
• Chapter IV : Make some predictions
• Chapter V : Deploy the model as a Web service
16. Chapter V :: Deploy your model as a Web Service
A. Sauvage
Now that the model have been built you can access it
through the web service hosted on the second
docker container we have built.
And so the forecast API can be used by other services
to do energy optimization, smart grid load balancing
strategies etc …
17. FIXER is here to fix your challenges !
Thank you.
Contact :: antoine.sauvage@fixer.co.jp
Want to know more about how to deploy your
machine learning developments on a managed
Cloud infrastructure ?