Deploying models to Real-time systems¶

How do I use my model in an app?¶

• Your external apps can't justr run a notebook
• Separate your training from your predictions
  • Train the model periodically offline
  • Push the model - or it's results - to a web service
  • Your app calls the web service

Example: Google Cloud ML¶

• Dump your trained classifier using sklearn.externals
  • from sklearn.externals import joblib
  • joblib.dump(clf, 'model.joblib')
• Upload model.joblib to Google Cloud storage, specifying the scikit-learn framework
• Cloud ML Engine exposes ar REST API that you call to make predictions in real-time

Example: AWS (recommender system)¶

• Server logs ->
• Amazon Kinesis Data Firehose ->
• Amazon S3 ->
• Amazon EMR ->
• Amazon DynamoDB ->
• AWS Lambda ->
• Amazon API Gateway ->
• Client app

Other approaches¶

• Roll your own web service with Flask or another framework
  • Then you have servers to provision and maintain.
• Go all-in with a platform
  • Amazon SageMaker
  • Amazon Comprehend
  • Amazon Lex
  • Amazon Polly
  • Amazon Rekognition
  • Amazon Rekognition Image
  • Amazon Rekognition Video
  • Amazon Translate
  • Amazon Transcribe
  • AWS Deep Learning APIs
  • AWS DeepLens