Upgrade to Pro
— share decks privately, control downloads, hide ads and more …
Speaker Deck
Features
Speaker Deck
PRO
Sign in
Sign up for free
Search
Search
Practical DevOps for the busy data scientist
Search
Tania Allard
October 09, 2019
Programming
1
860
Practical DevOps for the busy data scientist
Tania Allard
October 09, 2019
Tweet
Share
More Decks by Tania Allard
See All by Tania Allard
2024_pydata_lndn.pdf
trallard
1
220
The RSE hiring and career progression pipelines: Top tips to navigate them efficiently
trallard
0
260
Mentored Sprints - 2023
trallard
0
230
Mentored Sprints 2022 - kickoff
trallard
3
280
Como participar en el mercado emergente del codigo abierto
trallard
4
300
El presente y futuro del computo cientifico con Python
trallard
0
260
Foss for fun and profit
trallard
3
340
Open source for fun and profit: rethinking the long road of sustainability.
trallard
0
190
Docker and Python: making them play nicely and securely for Ml and DS
trallard
1
640
Other Decks in Programming
See All in Programming
Flutter × Firebase Genkit で加速する生成 AI アプリ開発
coborinai
0
160
ソフトウェアエンジニアの成長
masuda220
PRO
12
2k
GitHub Actions × RAGでコードレビューの検証の結果
sho_000
0
270
Formの複雑さに立ち向かう
bmthd
1
870
Serverless Rust: Your Low-Risk Entry Point to Rust in Production (and the benefits are huge)
lmammino
1
130
データの整合性を保つ非同期処理アーキテクチャパターン / Async Architecture Patterns
mokuo
51
17k
Amazon ECS とマイクロサービスから考えるシステム構成
hiyanger
2
570
AWS Organizations で実現する、 マルチ AWS アカウントのルートユーザー管理からの脱却
atpons
0
150
Rails アプリ地図考 Flush Cut
makicamel
1
120
.NET Frameworkでも汎用ホストが使いたい!
tomokusaba
0
110
PHPカンファレンス名古屋2025 タスク分解の試行錯誤〜レビュー負荷を下げるために〜
soichi
1
410
How mixi2 Uses TiDB for SNS Scalability and Performance
kanmo
39
15k
Featured
See All Featured
Building Your Own Lightsaber
phodgson
104
6.2k
KATA
mclloyd
29
14k
The Cult of Friendly URLs
andyhume
78
6.2k
The Invisible Side of Design
smashingmag
299
50k
Building Applications with DynamoDB
mza
93
6.2k
Bash Introduction
62gerente
611
210k
Measuring & Analyzing Core Web Vitals
bluesmoon
6
240
Adopting Sorbet at Scale
ufuk
74
9.2k
Keith and Marios Guide to Fast Websites
keithpitt
411
22k
Testing 201, or: Great Expectations
jmmastey
42
7.2k
Six Lessons from altMBA
skipperchong
27
3.6k
The Psychology of Web Performance [Beyond Tellerrand 2023]
tammyeverts
46
2.3k
Transcript
Practical DevOps for the busy data Scientist
bit.ly/PyConDE-mlops Slides
What you’ll learn 01 02 Why MLOps/ DevOps ? Who
is responsible? 03 04 Getting started Getting from A to B
About Me
Software engineering Algorithm Data Answers @ixek bit.ly/PyConDE-mlops
Machine learning Answers Data Algorithm @ixek bit.ly/PyConDE-mlops
Machine learning Answers Data Model @ixek bit.ly/PyConDE-mlops @ixek bit.ly/PyConDE-mlops
Machine learning Answers Data Model Answers Predictions @ixek bit.ly/PyConDE-mlops
The data cycle Magic? R&D Generation @ixek bit.ly/PyConDE-mlops
Anyone? @ixek bit.ly/PyConDE-mlops
A common scenario @ixek bit.ly/PyConDE-mlops
@ixek bit.ly/PyConDE-mlops
If you had one wish? @ixek bit.ly/PyConDE-mlops
Replacing the magic ML Ops and robust pipelines R&D Generation
@ixek bit.ly/PyConDE-mlops
How skills are perceived @ixek bit.ly/PyConDE-mlops
Better @ixek bit.ly/PyConDE-mlops
How they really are @ixek bit.ly/PyConDE-mlops
DevOps is the union of people, process, and products to
enable continuous delivery of value into production - Donovan Brown What is devops @ixek bit.ly/PyConDE-mlops
MlOps Aims to reduce the end-to-end cycle time and friction
of data analytics/science from the origin of ideas to the creation of data artifacts. What is devops @ixek bit.ly/PyConDE-mlops
But I do not work in a big company with
many ML engineers @ixek bit.ly/PyConDE-mlops
Build your own MLOps Platform @ixek bit.ly/PyConDE-mlops
None
None
Practical steps @ixek bit.ly/PyConDE-mlops
We have the notebooks in source control @ixek bit.ly/PyConDE-mlops
Your saviour Source control • Code and comments only (not
Jupyter output) • Plus every part of the pipeline • And Infrastructure and dependencies • And maybe a subset of data @ixek bit.ly/PyConDE-mlops
Everything should be in source control!! Except your training data
which should be a known, shared data source Do not touch the raw data! Not even with a stick Your saviour @ixek bit.ly/PyConDE-mlops
Deterministic environments @ixek bit.ly/PyConDE-mlops
Whatever that environment is @ixek bit.ly/PyConDE-mlops
Your laptop is not a production environment… so ensure reproducibility
@ixek bit.ly/PyConDE-mlops
@ixek bit.ly/PyConDE-mlops
Use pipelines for repeatability and reproducibility @ixek bit.ly/PyConDE-mlops
ml.azure.com
@ixek bit.ly/PyConDE-mlops
@ixek bit.ly/PyConDE-mlops
Automate wisely @ixek bit.ly/PyConDE-mlops
Adopt automation • Orchestration for Continuous Integration and Continuous Delivery
• Gates, tasks, and processes for quality • Integration with other services • Triggers on code and non-code events @ixek bit.ly/PyConDE-mlops
Complete pipeline @ixek bit.ly/PyConDE-mlops
Kubeflow example https://www.kubeflow.org/docs/azure/azureendtoend/ @ixek bit.ly/PyConDE-mlops
Build pipeline- https://azure.microsoft.com/en-us/services/devops/https://azure.microsoft.com/e n-us/services/devops/
Code event trigger @ixek bit.ly/PyConDE-mlops
Release / deploy @ixek bit.ly/PyConDE-mlops
In brief Deterministic environments Use pipelines Continuous integration and delivery
Source control (done right) Code, infrastructure, everything! Ensure production readiness For repeatable workflows Detect errors early and seamless deployments @ixek bit.ly/PyConDE-mlops
Want to learn more? • ml.azure.com • https://azure.microsoft.com/en-us/services/devops/ • https://docs.microsoft.com/en-us/azure/machine-learning/ser
vice/concept-ml-pipelines @ixek bit.ly/PyConDE-mlops
Come talk to us! @ ixek trallard@bitsandchips.me