Serverless for Data Scientists

on YouTube ↗46m

" summary

Claim: serverless removes OS management so data science runs as short-lived functions. Demonstrations use AWS Lambda with deployment via Zappa and parallel mapping with pi r n. Surprises: an Instapaper migration drops $10,000/mo to $370/mo, a embarrassingly parallel Black Scholes run shows ~270× speedup, and cold-starts add ~20s overhead. Built artifacts: cron bots and a Flask model server that downloads a pickled classifier from S3 and works within Lambda limits (≈1.5 GB RAM, 15 min timeout).

AI-written summary · about

" chapters (14)

1. 0:00Introduction
2. 1:09Move to cloud and serverless
3. 3:58What is serverless
4. 7:29Deploy web app serverless
5. 11:19Serverless cron jobs
6. 14:51Advantages and disadvantages of serverless
7. 17:43Data science use cases
8. 20:31Introducing PyRone
9. 24:45Embarrassingly parallel
10. 26:08Real-world examples
11. 30:18Hyperparameter optimization
12. 35:56Summary
13. 39:45When serverless fails
14. 40:51Q&A

" related talks

• 01h 00mServerless - reality or BS - notes from the trenchesLynn Langit · NDC Oslo 2017
• 40mThe State of Serverless ComputingChenggang Wu, Vikram Sreekanti · Craft Conference 2019
• 47mAdvanced Serverless Architectural Patterns on AWSAlex Casalboni · Heapcon 2019
• 54mServerless beyond FunctionsDanilo Poccia · Voxxed Days Bucharest 2018
• 02h 37mServerless Deep DiveBert Ertman · Devoxx Belgium 2017
• 24mServerless gives you wingsYan Cui · Frontend Love 2020
• 42mNo Server Is Better Than No Server: Building Blocks for Serverless ArchitecturesConstantin Gonzalez · code.talks 2017
• 51mServerless - a backend thing that gives superpowers to frontend developersSlobodan Stojanović · <React.NotAConf /> 2019