Public | Automated Build

Last pushed: 6 days ago
Short Description
Experiments
Full Description

ZooniverseExperimentServer

Implementation of planout for zooniverse for A/B splits and logging.

Experiments defined in /experiments.

Use the server by hitting up the following routes. All replies in JSON

  • GET /active_experiments?project= lists active experiments with an optional project param to limit it to experiments on a given project
  • GET /experiment/:experiment_name?userid=&pageid= gives back the params for an experiment given userid pageid etc. These params can be anything you want to select on, could be classification count etc. and should match what is defined in the experiment file.
  • POST /experiment/:experiment_name/log/:log_type will log all params to the experiment under the label log_type.

Note that :experiment_name should be replaced (without the colon) by one of the values returned from active_experiments.

New experiments will be accepted using pull requests.

#Intervention API

While planout provides an elegant way to run static A/B splits, often what we need is an intervention based on external models or data. The intervention API allows third parties to register interventions to be taken for a given user.

  • GET /users/:user_id/interventions will return all active interventions for a given user, except those for experiments which the user has been opted out of (these are represented by the suppressed_interventions count which is also returned, along with the user's opt out settings)
  • POST /users/:user_id/interventions will register an intervention for a user (see later for intervention format)
  • GET /interventions/:intervention_id allows you to poll the status of an intervention. It also shows the user's current opt out settings for this experiment.
  • POST /interventions/:intervention_id/detected allows you to mark an intervention as detected (provided the user is not opted out of that experiment)
  • POST /interventions/:intervention_id/delivered allows you to mark an intervention as delivered (provided the user is not opted out of that experiment)
  • POST /interventions/:intervention_id/dismissed allows you to mark an intervention as dismissed (provided the user is not opted out of that experiment)
  • POST /interventions/:intervention_id/completed allows you to mark an intervention as completed (provided the user is not opted out of that experiment)
  • GET /users/:user_id/optout will return the current opt out settings for the user for this experiment (project and experiment_name must be specified)
  • POST /users/:user_id/optout will opt out the user from the specified experiment (project and experiment_name must be specified). This means that any interventions posted to this experiment for this user will be suppressed and not returned or modifiable by the API.

To register an intervention you need to send a POST request with the following data to the server:

{
  "project":"galaxy_zoo",
  "experiment_name":"name of experiment",
  "cohort_id":1,
  "preconfigured_id":1,
  "state": "active"
  "intervention_type":"prompt user about talk",
  "text_message":"please return",
  "time_duration":60,
  "presentation_duration":20,
  "intervention_channel": "web model",
  "take_action" : "after_next_classification",
}
  • project : is the project on which the intervention should be taken
  • intervention_type: the name of the intervention
  • preconfigured_id: the id of the message to display (currently 1, 2 or 3 only)
  • experiment_name: the name of the experiment
  • text_message : an optional text message to send to the user
  • cohort_id : used along with a planout plan to select additional split properties for the user
  • time_duration: how long does the intervention remain active (in seconds). If the user does not see the intervention in this time the intervention is retired server-side before any client can see it (moves from active to inactive state)
  • presentation_duration: how long does the message stay on screen (in seconds)
  • intervention_channel : how should the intervention be delivered
  • take_action : at what point in a user's experience should the intervention be presented
  • state: the state of the intervention: initially active, until the first GET after time_duration has passed, at which point it becomes inactive. An active intervention will be set to detected when the client reports detection, and delivered when the client reports delivery. Additionally a delivered intervention may be set to dismissed if the user dismissed it before the presentation_duration has elapsed, and the client can set it to completed once the presentation_duration has elapsed.

On POSTing an intervention the API will return that intervention along with an ID which allows researchers to poll for the current state of the intervention. It will also return information about the current user's opt out status, and a value showing whether or not the user's current opt out status will cause this intervention to be suppressed.

#Example endpoints

Create intervention

See here for an example of creating an intervention

Interventions for a user

example

Poll interaction

example

Docker Pull Command
Owner
zooniverse
Source Repository

Comments (0)