Label Studio ML Backend

1.0.9 verified Mon Apr 27 auth: no python

Label Studio ML backend library for building, training, and deploying machine learning models with Label Studio. Version 1.0.9 supports Python >=3.6, released irregularly. Integrates with Label Studio for interactive labeling with model predictions.

pip install label-studio-ml

Common errors

error ModuleNotFoundError: No module named 'label_studio_ml' ↓

cause Package not installed or installed as 'label-studio-ml' but imported with wrong name.

fix

Run pip install label-studio-ml and import as 'label_studio_ml'.

error TypeError: predict() missing 1 required positional argument: 'tasks' ↓

cause Method signature mismatch: predict(tasks, **kwargs) is required.

fix

Define predict(self, tasks, **kwargs) returning list of dicts.

Warnings

gotcha LabelStudioMLBase is the correct base class; avoid custom inheritance from sklearn-like classes directly. Predictions must be a list of dicts with 'result' key. ↓

fix Ensure predict() returns a list of dictionaries, each containing 'result' (list of annotation results).

deprecated The old config system using label_config.xml is deprecated; use XML or JSON config from project settings. ↓

fix Use Label Studio web interface to generate config, or pass a LabelConfig string via label_config in init.

Imports

LabelStudioMLBase

wrong

from label_studio.ml import LabelStudioMLBase

correct

from label_studio_ml.model import LabelStudioMLBase

Incorrect import path from older docs or typos.

get_local_path
```
from label_studio_ml.utils import get_local_path
```
Utility function to resolve local file paths from data URIs.

Quickstart

Basic skeleton for a Label Studio ML backend model.

from label_studio_ml.model import LabelStudioMLBase
from label_studio_ml.utils import get_local_path

class MyModel(LabelStudioMLBase):
    def predict(self, tasks, context, **kwargs):
        return [{"result": []}]

    def fit(self, completions, workdir=None, **kwargs):
        pass