Yatta

Yet AnoTher Tool for Annotation

Yatta is a flexible framework for creating annotation tools that abstracts away the boring bits (user and data management) so that you can focus on quickly creating complex annotation interfaces.

Installation

Yatta is still in development, so it is not published on PyPI right now. However, you can install the latest code directly from Github or the latest release from the test PyPI server:

pip install git+https://github.com/naitian/yatta

or

pip install -i https://test.pypi.org/simple/ yatta-tool

Quickstart

Your Yatta app is as simple as a single Python file.

from yatta.core import Yatta
from yatta.web import Server
from yatta.distributor import AllDistributor
from yatta.ordering import SequentialOrdering
from yatta.base import Textbox

yatta = Yatta(
    dataset=[
        {"text": "Hello, world!"},
        {"text": "Goodbye, world!"},
        {"text": "Hello, again!"},
        {"text": "Goodbye, again!"},
        {"text": "Hello, one more time!"},
    ],
    task={"text": Textbox()},
    distributor=AllDistributor,
    ordering=SequentialOrdering,
)

with yatta.session():
    yatta.assign_tasks()
    yatta.assign_all_orderings()

server = Server(yatta, port=5000, secret_key="very_secret_key")
server.run()

Let's break this down. A Yatta app consists of the following configurable pieces:

1. The dataset is any sequence with a __getitem__ and __len__. This means you can even directly use Torch datasets.
2. The task is a dictionary of components. Each component includes front-end code that defines how data are rendered in the annotation interface.
3. The distributor determines how items in the dataset are allocated between users.
4. The ordering determines the order in which the user encounters the data during the annotation process.

The Yatta app itself exposes functions to manage annotations and users. To use the web interface, you can wrap the Yatta app with a Server and call server.run().