Automating the boring task of submitting travel expenses we developed ML model for classifying recipes. Using AWS EC2, Lambda, S3, SageMaker, Rekognition we evaluated different ways of training model and serving predictions as well as different model approaches (classical ML vs. Deep Learning).
Raspberry Pi 5: Challenges and Solutions in Bringing up an OpenGL/Vulkan Driv...
Hotel or Taxi? "Sorting hat" for travel expenses with AWS ML infrastructure
1. Hotel or Taxi?
"Sorting hat" for travel expenses
with AWS ML infrastructure.
BERLIN, 18.OCT 2018
MICHAEL PERLIN
2. www.innoq.com
SERVICES
Strategy & technology consulting
Digital business models
Software architecture & development
Digital platforms & infrastructures
Knowledge transfer, coaching & trainings
Big data & machine learning
FACTS
~130 employees
Privately owned
Vendor-independent
OFFICES
Monheim
Berlin
Offenbach
Munich
Hamburg
Zurich
CLIENTS
Finance
Telecommunications
Logistics
E-Commerce
Fortune 500
SMBs
Startups
3. Agenda
• The value of machine learning
• Problem we‘ve solved
• AWS infrastructure for training
• How deep learning works
• How we run it in production
21. Training environment
• EC2 Instance with bare Linux
• Install libraries
• Configure GPU usage
• Install Jupyter
• Add self-signed certificates
• Go!
Option 1
22. Training environment
• EC2 Instance with AMI from
Marketplace containing pre-
installed and pre-configured
libraries
• Add self-signed certificates
• Go!
Option 2
40. Training
• With all the data iterate until error stops to shrink
• The result of the adjustments is the trained
model
• Now it can be deployed into production
47. Inference
• General approach: load the model saved by
training, feed the input, get output
• Even cross-language works, i.e. model trained
with Python can be used a Java application
• Usually works on commodity hardware
48. Package, Build and Deploy
web framework
docker containerOption 1
container scheduler of your choice: EKS, ECS,
OpenShift, Giant Swarm...
deploy
• inference code
• trained model
• dependent libs
49. Package, Build and Deploy
• inference code
• trained model
• dependent libs
Option 2
deploy
zip
AWS Lambda