Skip to content

An AB test server which can run all of you AB tests with minimal configuration and uses the epsilon greedy algorithm to optionally auto optimise the tests.

License

Notifications You must be signed in to change notification settings

kishoreyekkanti/epsilon-ab

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

22 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Features

  1. Setup as many tests as you want with as many branches(options) as you want, using simple REST api
  2. Get which branch to be shown to a given user by providing some user unique id. Once the option is determined then that option is pinned to that user so future requests for that user unique id will always generate the same branch/option
  3. Fetch a single test detail by id (or) all the tests setup in the system
  4. Update the branch/option details on the fly if you want to add more weightage to any specific branch
  5. Reward a branch and update the conversion funnel for any given branch using a simple API
  6. Gather the basic stats about how your AB tests are performing using an API
  7. You can always gather the complex metrics by directly querying the database
  8. Supports Postgresql and Mysql
  9. Runs on Node.js and PM2. So you need not bother about the app being down as PM2 can take care of automatic restarts if in case the app dies
  10. Json based logging using Bunyan
  11. A postman collection of all the apis is attached at the bottom of this documentation, you can download and play with it
  12. This project uses the strategy explained by Steve Hanov in his blog

Usage

Installation

Download

Clone the project and run npm install to download all the dependencies.

Configuration

Before you start have a look at config/config.json and update the default hash appropriately. Don't forget to update the name key of the default hash as it is important to pick up the database type. Other keys in the config file are ignored. Example postgresql configuration can be as below. "dev" just resembles its being run on dev mode. You can change it to "production" (if you want to) and pass in your NODE_ENV variable at the start up time.

Postgresql configuration (config/config.json)

Change the db credentials as required. Below is just an example.

{
  "dev": {
      "default": {
        "host": "localhost",
        "port": "5432",
        "database": "eps",
        "user": "epsuser",
        "password": "epsuser",
        "name": "pgsql"
      }
    }
}
Mysql configuration (config/config.json)

You need the mysql.sock socketPath if you are on Mac. Change the db credentials as required. Below is just an example.

{
  "dev": {
      "default": {
      "name": "mysql",
      "host": "localhost",
      "port": "3306",
      "database": "eps",
      "user": "root",
      "password": "root",
      "connectionLimit": 5,
      "socketPath": "/Applications/MAMP/tmp/mysql/mysql.sock"
      }
    }
}
Table creation

Depending on the above configuration you have to create the tables in your database. Table and index creations are under schema/pgsql and schema/mysql for postgres and Mysql respectively.

Running

If you want to test this in your local machine and if you don't bother about PM2 then you can just run node app.js within the project root directory and the application will start by default on port 5000.

If you want to run this using PM2 then you can use the command pm2 start app.js -i 1 -n epsilon-ab . This will start one instance of the server with the name epsilon-ab. You can use pm2 list or pm2 show epsilon-ab to see the details of the process. To stop the app use pm2 stop epsilon-ab. To understand different options of PM2 use this link.

Logging

Once the app starts you should see the logs flowing in json format. Bunyan is used for logging. Have a look at Bunyan to understand the best way of grepping etc,.

APIs

Create a test with 3 branches(options)

Provided the server is running on port 5000 - use the following curl command to create a test. auto_optimise is a really powerful feature if once set its automatically determines the best option based on the user behaviour and slowly becomes the default behaviour for the entire test. Each branch you want to create will have a unique option_no starting from 1. weightage is the weight you want to attach to each branch. The total wieghtage of all the branches need NOT sum up to 100. weightage is more relative than a percentage (though you can use it as a percentage). status can take active or inactive. test_description is a free flowing text to help you identify the purpose of the branch. test_name should be unique across different tests.

REQUEST

curl -X POST -H "Content-Type: application/json" -H "Cache-Control: no-cache"  -d '{
    "test_name":"show_button_colors_for_signup",
    "options":[
        {
            "option_no":1,
            "weightage":30,
            "auto_optimise":true,
            "status":"active",
            "test_description":"Show Red Color"
        },
        {
            "option_no":2,
            "weightage":40,
            "auto_optimise":true,
            "status":"active",
            "test_description":"Show Blue Color"
        },
        {
            "option_no":3,
            "weightage":30,
            "auto_optimise":true,
            "status":"active",
            "test_description":"Show Green Color"
        }
        
        ]
}' "http://localhost:5000/epsTest"

RESPONSE

{
  "response": "Saved the AB tests successfully"
}
Get an option for a user

Once the tests are setup you want to get the branch selected by the algorithm for a user. Here user_unique_id can be a cookie or customer id or any id which you can uniquely identify the user visit with. And as mentioned above, once the option is generated, its pinned for that particular user_unique_id and the same option will be given back for subsequent calls.

REQUEST

curl -X GET -H "Content-Type: application/json" -H "Cache-Control: no-cache"  "http://localhost:5000/epsOption?user_unique_id=aasd2345&test_name=show_beta_navigation"

RESPONSE So the option/branch selected by the algorithm is 3

{
  "testName": "show_beta_navigation",
  "option": 3,
  "createdAt": "2016-06-09T21:53:52.234Z"
}
Update a test option

If you want to change the test description or change the weightage of any option(this will skew the results as the test was already setup), you can use this api.

REQUEST

curl -X POST -H "Content-Type: application/json" -H "Cache-Control: no-cache"  -d '{
    "options":{
        "weightage":30,
        "auto_optimise":false,
        "status":"active",
        "test_description":"updated the Blue color option"
    }
}' "http://localhost:5000/epsTest/20/options/2"

RESPONSE

{
  "response": "Eps option updated successfully"
}
Award an option

Say in the above example of Red, Blue and Green buttons if the user selected Red button and you want to reward that selection of user then you should use this api.

REQUEST

curl -X PUT -H "Content-Type: application/json" -H "Cache-Control: no-cache"  -d '{
    "user_unique_id":"aasd2345"
}' "http://localhost:5000/epsTest/reward"

RESPONSE

{
  "response": "reward updated"
}
Update conversion

If a user has chosen a branch and completed a desired funnel, then you can mark a conversion for the given branch. Ex: If a user choses a red button and signs up, checks out and pays successfully, then you can mark a conversion for red button branch.

This conversion updates will really help to determine the quality of traffic for a given ab test branch. Ex: More people click on Red button but the conversion is low, where as less people click on green button with huge conversions etc,. Along with this request you can also optionally send the user_domain_id. This is nothing but the other user_unique id but in domain terms. Say for example during the initial branch selection we send the cookie string as unique id, post signup we might have got the real user id of the user and we want to update the real user id(user_domain_id against the cookie id.

REQUEST

curl -X PUT -H "Content-Type: application/json" -H "Cache-Control: no-cache" -d '{
    "user_unique_id":"aasd2345",
    "user_domain_id":"12345"
}' "http://localhost:5000/epsTest/conversion"

RESPONSE

{
  "response": "conversion updated"
}
Get All AB Tests

To fetch all the ab tests setup in the system

REQUEST

curl -X GET -H "Content-Type: application/json" -H "Cache-Control: no-cache" "http://localhost:5000/epsTest"

RESPONSE

[
  {
    "id": 40,
    "test_name": "show_button_coslors_for_signup",
    "test_description": "Show Green Color",
    "option_no": 3,
    "weightage": 30,
    "auto_optimise": true,
    "status": "active",
    "created_at": "2016-06-13T23:23:21.000Z",
    "updated_at": null
  },
  {
    "id": 39,
    "test_name": "show_button_coslors_for_signup",
    "test_description": "Show Blue Color",
    "option_no": 2,
    "weightage": 40,
    "auto_optimise": true,
    "status": "active",
    "created_at": "2016-06-13T23:23:20.988Z",
    "updated_at": null
  },
  {
    "id": 38,
    "test_name": "show_button_coslors_for_signup",
    "test_description": "Show Red Color",
    "option_no": 1,
    "weightage": 30,
    "auto_optimise": true,
    "status": "active",
    "created_at": "2016-06-13T23:23:20.943Z",
    "updated_at": null
  },
  {
    "id": 37,
    "test_name": "show_button_colors_for_signup2",
    "test_description": "Show Green Color",
    "option_no": 3,
    "weightage": 30,
    "auto_optimise": true,
    "status": "active",
    "created_at": "2016-06-10T19:45:05.920Z",
    "updated_at": null
  },
  {
    "id": 36,
    "test_name": "show_button_colors_for_signup2",
    "test_description": "Show Blue Color",
    "option_no": 2,
    "weightage": 40,
    "auto_optimise": true,
    "status": "active",
    "created_at": "2016-06-10T19:45:05.911Z",
    "updated_at": null
  },
  {
    "id": 35,
    "test_name": "show_button_colors_for_signup2",
    "test_description": "Show Red Color",
    "option_no": 1,
    "weightage": 30,
    "auto_optimise": true,
    "status": "active",
    "created_at": "2016-06-10T19:45:05.897Z",
    "updated_at": null
  },
  {
    "id": 33,
    "test_name": "show_button_colors_for_signup1",
    "test_description": "Show Green Color",
    "option_no": 3,
    "weightage": 30,
    "auto_optimise": true,
    "status": "active",
    "created_at": "2016-06-10T19:40:38.589Z",
    "updated_at": null
  },
  {
    "id": 32,
    "test_name": "show_button_colors_for_signup1",
    "test_description": "Show Blue Color",
    "option_no": 2,
    "weightage": 40,
    "auto_optimise": true,
    "status": "active",
    "created_at": "2016-06-10T19:40:38.570Z",
    "updated_at": null
  },
  {
    "id": 31,
    "test_name": "show_button_colors_for_signup1",
    "test_description": "Show Red Color",
    "option_no": 1,
    "weightage": 30,
    "auto_optimise": true,
    "status": "active",
    "created_at": "2016-06-10T19:40:38.533Z",
    "updated_at": null
  }
]
Get AB tests by id

Fetch AB test by an id

REQUEST

curl -X GET -H "Content-Type: application/json" -H "Cache-Control: no-cache"  "http://localhost:5000/epsTest/21"

RESPONSE

{
  "id": 21,
  "test_name": "show_button_colors_for_signup",
  "test_description": "Show Green Color",
  "option_no": 3,
  "weightage": 30,
  "auto_optimise": true,
  "status": "active",
  "created_at": "2016-06-09T21:35:00.725Z",
  "updated_at": null
}
Get stats on how the tests are performing

Fetches the performance of the given AB test. Trails are the number of times the option is given across different users. Reward is how many number of times user used the given option/branch. Ex: Out of 30 times(trails) only 3 times(rewards) users clicked on the red button. conversion shows how many times the funnel conversion happened for a given test branch.

REQUEST

curl -X GET -H "Content-Type: application/json" -H "Cookie: PHPSESSID= fd1c94418019fc04f77b63ed5ae8cf1f" -H "Cache-Control: no-cache" "http://localhost:5000/epsTest/show_beta_navigation/stats"

RESPONSE

{
  "test_name": "show_beta_navigation",
  "options": [
    {
      "option_no": 2,
      "trial": 40,
      "reward": 10,
      "weightage": 25,
      "auto_optimise": true,
      "status": "active",
      "test_description": "Shows beta navigation with optional links",
      "conversion": "10"
    },
    {
      "option_no": 3,
      "trial": 80,
      "reward": 75,
      "weightage": 25,
      "auto_optimise": true,
      "status": "active",
      "test_description": "Shows beta navigation with no links",
      "conversion": "14"
    }
  ]
}

Postman collection of the API

Use this link to download the postman collection. You can directly import this collection and play with the api. Make sure you run the seed data under schema/*/seed.sql to have some dummy data to start with.

Future enhancements

  1. Generating more stats in terms of AB tests like send te time series data of how the ab tests performed over time
  2. Create a ORM like adapter across different databases so the sql generation will be easier.

About

An AB test server which can run all of you AB tests with minimal configuration and uses the epsilon greedy algorithm to optionally auto optimise the tests.

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published