MLServices

• reference from http://predictionio.incubator.apache.org/system/

Module structure (services)

Builder service

dataset

• prepare training data
• prepare entity id map

algorithm

• train model
• save model or just dump result

builder

• coordinate dataset to algorithm
• subscribe for BUILD_MODEL command
• publish NEW_MODEL for engine
• builder service app

API service

model

• load model or/and using dataset
• predict, recommend, or serve function

engine

• coordinate model for API service(serving)
• subscribe NEW_MODEL command

serving

• Flask API route
• API service app

App entry point

• app.py for builder or api

Example

Required

• pyspark (1.6 or 2.0)
• keras (tensorflow & theano) - contribute from RC

Services

• Iris classifier
• MoviesLens recommender

Run Microservice

• Start iris service on port 5001, movielens service on port 5002

  # should be execute the bin/start.sh firstly
  # 1. it will export the enviornment python path
  # 2. it will start the redis-server
  # 3. (****) The following commands, the working space should be 'ROOT FOLDER' of this repository

  # == IRIS DNN classifier ==
  python sbin/app.py --func mining --service iris
  python sbin/app.py --func api --service iris
  # redis-cli: PUBLISH iris_mining BUILD_MODEL
  # redis-cli: PUBLISH iris_api KILL

  # == MovieLens recommender ==
  python sbin/app.py --func mining --service movielens
  python sbin/app.py --func api --service movielens
  # redis-cli: PUBLISH movie_mining BUILD_MODEL
  # redis-cli: PUBLISH movie_api KILL

• Routes

1. http://localhost:5001/features/2,3,4,1/class
2. http://localhost:5001/features/2,3,4,1/probs
3. http://localhost:5002/users/1/top/10