Kili Tutorial: How to query using the API

In this tutorial, we will show query useful information through Kili's API, interacting directly with the database.

There are 6 different types of data you could be interested in querying, and it is higly customizable :

Your organization's info
The organization's users
KPI's and labeling data for different projectUsers
The whole project or selected parts
The assets themselves
Obviously, the labels

All those queries request, among other arguments, a skip and first argument, that control the number of results retrieved. skipis the rank at which the result start, and first is the extent of the query.

Additionally, for an overview of Kili, visit the website, you can also check out the Kili documentation, or some other recipes.

For more information, the architecture of our API is available here, all fields are precisely detailled and can be queried, except those related to authentication. Let's sign in, create a project, upload some assets and import predictions, if you need precisions on the steps carried-out below, you can refer to the mentionned recipes.

We now have a project ready, which information is in a project_example dictionnary, and which data we are going to query.

#!pip install kili 
from kili.authentication import KiliAuth
from kili.playground import Playground
import os

email = os.getenv('KILI_USER_EMAIL')
password = os.getenv('KILI_USER_PASSWORD')
api_endpoint = os.getenv('KILI_API_ENDPOINT')
kauth = KiliAuth(email=email, password=password, api_endpoint=api_endpoint)
playground = Playground(kauth)

project_example = {
    'title': 'Porsche or Tesla recognition',
    'description': 'Identify and locate cars',
    'input_type': 'IMAGE',
    'json_interface': {
        "jobs": {
            "JOB_0": {
                "mlTask": "OBJECT_DETECTION",
                "tools": [
                    "rectangle"
                ],
                "instruction": "What car brand ?",
                "required": 1,
                "isChild": False,
                "content": {
                    "categories": {
                        "TESLA": {"name": "Tesla"},
                        "FERRARI": {"name": "Ferrari"}
                    },
                    "input": "radio"
                }
            }
        }
    },

    'assets_to_import': [
        "https://images.caradisiac.com/logos/3/8/6/7/253867/S0-tesla-enregistre-d-importantes-pertes-au-premier-trimestre-175948.jpg",
        "https://img.sportauto.fr/news/2018/11/28/1533574/1920%7C1280%7Cc096243e5460db3e5e70c773.jpg"],

    'json_response': {
        "JOB_0": {
            "annotations": [{
                "boundingPoly": [{
                    "normalizedVertices": [
                        {"x": 0.16, "y": 0.82},
                        {"x": 0.16, "y": 0.32},
                        {"x": 0.82, "y": 0.32},
                        {"x": 0.82, "y": 0.82}
                    ]}
                ],
                "categories": [{"name": "TESLA", "confidence": 100}],
            }]
        }
    },
    'model_name': 'car-brand-localisation-v0.0.1'
}
project = playground.create_empty_project(user_id=playground.auth.user_id)
playground.update_properties_in_project(
    project_id=project['id'],
    title=project_example['title'],
    description=project_example['description'],
    input_type=project_example['input_type'],
    json_interface=project_example['json_interface'])
playground.append_many_to_dataset(
    project_id=project['id'],
    content_array=project_example['assets_to_import'],
    external_id_array=['ex1', 'ex2'])
playground.create_predictions(
    project_id=project['id'],
    external_id_array=['ex1'],
    model_name_array=[project_example['model_name']],
    json_response_array=[project_example['json_response']])

1. Organization info

We can query the information for the organization for specific users, and check that those parameters are correctly configured, either by providing

the id of the organization
a specific email

Another argument is the fields that need to be returned, that can be customized using our graphQL doc here

fields = ['id', 'name','address','country','zipCode','city']
playground.organizations(email=email, fields=fields, first=3, skip= 0)

2. User info

We can query the information for an user, and check that those parameters are correctly configured, either by providing

the email of the user
the organization_id of the users

fields =['id','activated','createdAt','email','name', 'organization.name', 'organization.city','organizationRole','updatedAt']
playground.users(email=email, fields=fields, skip=0, first=10)

3. ProjectUser info

We can query the information for a projectUser, and check that those parameters are correctly configured, either by providing

the email of the projectUser
the id of the organization
the id of the project
the id of the projectUser

fields=['id','activated','consensusMark',"honeypotMark","numberOfAnnotations","numberOfLabels",'role',
        'user.email', 'user.name', 'user.organization.name', 'user.organization.zipCode']
playground.project_users(project_id=project['id'], fields=fields, first=10, skip=0)

4. Project info

We can query the information for a project, and check that those parameters are correctly configured, either by providing

the email of the projectUser
a part of the title of the description search_query

Another argument is the fields that need to be returned, that can be customized using our graphQL doc here. Notice that the nested requests are possible, to retrieve the information at the projectUser's level, via the role subfield, and hence, once again, at the user or the organization's level. It is also possible to query the information on the author of the project, always with the same syntax :

fields = ['id','author.name','author.email', 'author.organization.name','consensusMark', 'createdAt',
          'description','honeypotMark', 'id', 'inputType', 'interfaceCategory', 'jsonInterface', 
          'maxWorkerCount', 'minConsensusSize', 'numberOfAssets', 'numberOfRemainingAssets','numberOfRoles', 
          'roles.id','roles.consensusMark', 'roles.numberOfAnnotations', 'roles.numberOfLabeledAssets', 
          'roles.numberOfLabels', 'roles.totalDuration', 'roles.user.email', 'roles.user.id', 'roles.user.name', 
          'title','titleAndDescription','useHoneyPot']
title = playground.projects(project_id=project['id'], fields=fields, first=10, skip=0)[0]['title']
playground.projects(project_id=project['id'], fields=fields, first=10, skip=0)

5. Labels info

We can query the information of labels, in order to use it later in our pipeline, either by providing

the idof the label or the asset to query
the id of the project
a part of the title or the description search_query
many other parameters to filter the list, the details can be found here, among which a part of the externalId of the asset, the status of the asset, the honeypot mark,etc...

fields = ['id','honeypotMark','numberOfAnnotations','jsonResponse','labelType','skipped','createdAt',
        'author.email', 'author.name', 'author.organization.name', 'author.organization.zipCode']
playground.labels(project_id=project['id'], fields=fields, first=10, skip=0)

6. Assets info

We can query the information of assets, along with the labels or more info, either by providing

the idof the asset to query
the id of the project
a part of the title or the description search_query
many other parameters to filter the list, the details can be found here, among which a part of the externalId, the status, the consensus level,etc...

fields=['id', 'honeypotMark', 'project.jsonInterface','content', 'createdAt', 'priority',
        'project.author.email', 'project.author.name', 'labels.jsonResponse', 'status',
        'labels.author.email', 'labels.author.organization.name', 'project.inputType']
playground.assets(project_id=project['id'], fields=fields, first=10, skip=0)

assert project_example['title'] == title