2. APPLYING DEVOPS PRACTICES TO THE DEVELOPMENT AND
DEPLOYMENT OF MACHINE LEARNING ALGORITHMS
WHAT IS MLOPS?
▸MLOPS is a combination of DevOps practices meets
Machine Learning.
▸The training, deployment and continued re-deployment of
ML models can be a costly and slow process without
employing some sort of automated and standardised
process into the mix.
▸Approximately only 10%-15% of ML models are ever put into
production so therefore a process to improve the reliability
and deliverability of these models is required.
3. APPLYING DEVOPS PRACTICES TO THE DEVELOPMENT AND
DEPLOYMENT OF MACHINE LEARNING ALGORITHMS
WHAT IS MLOPS?
▸“Companies are doing machine learning, but when you want
to scale them into production, that's where the rubber hits
the road. The hard part is putting it into production, which is
where companies have little experience. This gap I'm
describing really requires a bridge and this bridge is MLOps”
▸-Sivan Metzger, the CEO of ParallelM (a company
specialising specifically in MLOPS solutions)
4. APPLYING DEVOPS PRACTICES TO THE DEVELOPMENT AND
DEPLOYMENT OF MACHINE LEARNING ALGORITHMS
WHAT IS DEVOPS?
▸DevOps is an amalgamation of the words Development and
Operations. It bridges the gap between these two distinct
phases of the SDLC.
▸When successfully practiced, it results in reduced time to
deliver new software features or bug fixes as well as being
more responsive to changing business demands.
▸A 2017 research paper published revealed that there was
confusion in organisations over what DevOps is or should
be, and how best to quantifiably measure the results of it.
5. APPLYING DEVOPS PRACTICES TO THE DEVELOPMENT AND
DEPLOYMENT OF MACHINE LEARNING ALGORITHMS
WHY THE AMBIGUITY AROUND
DEVOPS?
▸It is both a cultural shift in how enterprises approach
software development as well as a technological change in
terms of tooling.
▸ To sum up DevOps in three terms it would be enhanced
communication, shared responsibility and increased
automation. The increased automation comes mostly from
the technology and tools being used, but the other two
attributes are mostly human driven.
6. APPLYING DEVOPS PRACTICES TO THE DEVELOPMENT AND
DEPLOYMENT OF MACHINE LEARNING ALGORITHMS
DEVOPS: CONVERGENCE OF
SOFTWARE TEAMS INTO ONE WHOLE
7. APPLYING DEVOPS PRACTICES TO THE DEVELOPMENT AND
DEPLOYMENT OF MACHINE LEARNING ALGORITHMS
DEVOPS - TECHNOLOGY MEETS
PEOPLE
▸The technology mostly consists of cloud based tools that allow
for better automation of mundane tasks involved with the
building, configuration, deployment and testing of software
throughout the SDLC process.
▸As for the human driven components of DevOps, it requires the
breaking down of traditional barriers between software teams.
Devs will start to think more like Ops personnel and vice versea.
This facilitates the fostering of shared responsibility between
teams and less victim blaming when something goes wrong in
the SDLC.
8. APPLYING DEVOPS PRACTICES TO THE DEVELOPMENT AND
DEPLOYMENT OF MACHINE LEARNING ALGORITHMS
WHAT IS MACHINE LEARNING?
▸ML is a data analytics technique that teaches computers to
do what comes naturally to humans which is to learn from
experience.
▸ML algorithms use computational methods to learn
information directly from data without relying on a
predetermined equation as a model.
▸The algorithms adaptively improve their performance as the
number of samples available for learning increases.
9. APPLYING DEVOPS PRACTICES TO THE DEVELOPMENT AND
DEPLOYMENT OF MACHINE LEARNING ALGORITHMS
WHAT IS MACHINE LEARNING?
10. APPLYING DEVOPS PRACTICES TO THE DEVELOPMENT AND
DEPLOYMENT OF MACHINE LEARNING ALGORITHMS
VALUE OF MACHINE LEARNING?
▸ML techniques when used in business applications can see
profits increase by up to 15% in recent studies by increasing
. But there is also a high failure rate too.
▸MLOPS is really about the democratising of Machine
Learning, bringing down the barrier for entry and allowing
data scientists to collaborate fully with IT Operators and
software developers in the same manner that DevOps
increased the reliability and decreased the deployment time
of regular software development.
11. APPLYING DEVOPS PRACTICES TO THE DEVELOPMENT AND
DEPLOYMENT OF MACHINE LEARNING ALGORITHMS
RESEARCH METHOD - AZURE MACHINE
LEARNING
▸The proposed research project will involve using Microsoft’s
suite of Machine Learning tools available on their Azure
Cloud platform. These have been available since 2015, but
only just in May 2019 have two new additions to the Azure
ML platform have made this an especially exciting platform.
12. APPLYING DEVOPS PRACTICES TO THE DEVELOPMENT AND
DEPLOYMENT OF MACHINE LEARNING ALGORITHMS
RESEARCH METHOD - AZURE MACHINE
LEARNING
▸The two new additions to Azure ML are:
▸Dedicated MLOPS capabilities built into the platform under
the product title of “DevOps For Machine Learning”. A
combination of Azure Pipelines + Azure DevOps with a
plugin to enable MLLOPs capabilities.
▸A visual drag and drop interface for modifying how datasets
are fed into ML models and deployed. No need for coding
knowledge required even though that is still an option for
more finer control.
13. APPLYING DEVOPS PRACTICES TO THE DEVELOPMENT AND
DEPLOYMENT OF MACHINE LEARNING ALGORITHMS
RESEARCH METHOD - AZURE MACHINE
LEARNING
▸Microsoft have some readymade ML projects developed with
MLOPS in mind to play around with first to familiarise myself with
their tools.
▸A custom ML application will be developed afterwards and
deployed using Azure MLOPS pipelines. Metrics are available in
Azure ML to see how well the newly deployed model is
performing.
▸The nature of the custom ML application and the business domain
it functions in has yet to be decided, but will be one for the first
things decided at the onset of the project
14. APPLYING DEVOPS PRACTICES TO THE DEVELOPMENT AND
DEPLOYMENT OF MACHINE LEARNING ALGORITHMS
RESEARCH METHOD - SURVEY
▸While conducting the literature review there was a research
paper on DevOps which conducted a survey of DevOps
professionals to try and pinpoint down exactly what DevOps
was and the value it brings.
▸A similar survey could be conducted as part of this project
with ML professionals to see if they see the value of what
MLOPS brings to ML and how best to incorporate it.
15. APPLYING DEVOPS PRACTICES TO THE DEVELOPMENT AND
DEPLOYMENT OF MACHINE LEARNING ALGORITHMS
RESEARCH METHOD - GANTT CHART
OF PROJECT PROGRESSION