Kili Python SDK

SDK reference

What is Kili?

Kili is a platform that empowers a data-centric approach to Machine Learning through quality training data creation. It provides collaborative data annotation tools and APIs that enable quick iterations between reliable dataset building and model training. More info here.

Annotation tools examples

Named Entities Extraction and Relation	PDF classification and bounding-box	Object detection (bounding-box)

and many more.

What is Kili Python SDK?

Kili Python SDK is the Python client for the Kili platform. It allows to query and manipulate the main entities available in Kili, like projects, assets, labels, api keys... It comes with several recipes that demonstrate how to use it in the most frequent use cases.

Requirements

• Python >= 3.7
• Create and copy a Kili API key
• Add the KILI_API_KEY variable in your bash environment (or in the settings of your favorite IDE) by pasting the API key value you copied above:

export KILI_API_KEY='<your api key value here>'

Installation

Install the Kili client with pip:

pip install kili

Usage

Instantiate the Kili client:

from kili.client import Kili
kili = Kili()
# You can now use the Kili client!

Note that you can also pass the API key as an argument of the Kili initialization:

kili = Kili(api_key='<your api key value here>')

Here is a sample of the operations you can do with the Kili client:

Creating an annotation project

json_interface = {
    "jobs": {
        "CLASSIFICATION_JOB": {
            "mlTask": "CLASSIFICATION",
            "content": {
                "categories": {
                    "RED": {"name": "Red"},
                    "BLACK": {"name": "Black"},
                    "WHITE": {"name": "White"},
                    "GREY": {"name": "Grey"}},
                "input": "radio"
            },
            "instruction": "Color"
        }
    }
}
project_id = kili.create_project(
        title="Color classification",
        description="Project ",
        input_type="IMAGE",
        json_interface=json_interface
)["id"]

Importing data to annotate

assets = [
    {
        "externalId": "example 1",
        "content": "https://images.caradisiac.com/logos/3/8/6/7/253867/S0-tesla-enregistre-d-importantes-pertes-au-premier-trimestre-175948.jpg",
    },
    {
        "externalId": "example 2",
        "content": "https://img.sportauto.fr/news/2018/11/28/1533574/1920%7C1280%7Cc096243e5460db3e5e70c773.jpg",
    },
    {
        "externalId": "example 3",
        "content": "./recipes/img/man_on_a_bike.jpeg",
    },
]

external_id_array = [a.get("externalId") for a in assets]
content_array = [a.get("content") for a in assets]

kili.append_many_to_dataset(
        project_id=project_id,
        content_array=content_array,
        external_id_array=external_id_array,
)

See the detailled example in this notebook.

Importing predictions

prediction_examples = [
    {
        "external_id": "example 1",
        "json_response": {
            "CLASSIFICATION_JOB": {
                "categories": [{"name": "GREY", "confidence": 46}]
            }
        },
    },
    {
        "external_id": "example 2",
        "json_response": {
            "CLASSIFICATION_JOB": {
                "categories": [{"name": "WHITE", "confidence": 89}]
            }
        },
    }
]

kili.create_predictions(
        project_id=project_id,
        external_id_array=[p["external_id"] for p in prediction_examples],
        model_name_array=["My SOTA model"] * len(prediction_examples),
        json_response_array=[p["json_response"] for p in prediction_examples])

See detailled examples in this notebook.

Exporting labels

assets = kili.assets(project_id=project_id)

with open("labels.json", "w") as label_file:
    for asset in assets:
        for label in asset.labels:
            label_file.write(label.json_response(format='simple'))

See a detailled example in this notebook.

More examples

You can find all several recipes in this folder. Among them:

• How to import assets (run it here)
• How to export labels (run it here)
• How to import predictions (run it here)
• How to query data through the API (run it here)

For more details, read the SDK reference or the Kili documentation