A server running multi-armed-bandit algorithm. A redis server is required to implement caching and delayed learning.

• As a consumer application,
• I want to know which feature to display,
• In order to display that feature.

• As an analytic guy,
• I want to get the feedback on the feature displayed,
• In order to improve the feature selection mechanism.

• As an analytic guy,
• I want to view the current stats,
• In order to know how the features perform.

$ curl http://localhost:9090/arms

Output:

{
  "arm":2,
  "id":"bcbl5jle4b3ipsuela90",
  "created_at":"2018-06-06T03:19:42.336425107Z",
  "updated_at":"2018-06-06T03:19:42.33642513Z"
}

$ curl -X POST -d {"arm": 2, "id":"bcbl5jle4b3ipsuela90", "reward": 1} http://localhost:9090/arms

Output:

{"ok": true}

$ curl http://localhost:9090/arms/stats

Output:

{
  "counts":[0,0,1],
  "rewards":[0,0,1],
  "experiment_name":"colors",
  "features":["red","green","blue"]
}

• dockerize the service
• create a dashboard for the service (web ui)
• make the service configurable
• make the results of the service transparent (like feature X is n% better than feature Y)
• isolate configuration
• add labels and annotations for registry
• create sdk to integrate the functionality into other applications
• or make this into a sidecar proxy
• CRUD and API interface and CLI to integrate with the configuration
• fix the race issue that is found in the api
• load data from persistent store (e.g. redis)
• note that features might not be loaded in the correct order, hence it is important not to hardcode the values. use namespace instead