Backend

Getting Started

This is an example of the backend that can be used with the React frontend part. By default, the backend uses label tool scripts from ../build.

Linux & Ubuntu guide

1 Install python and virtualenv. We recommend to use python 3.6, but probably python 2 solution will work too.

apt install python3.6
pip3 install virtualenv

2 Setup python virtual environment

virtualenv -p python3 env3
source env3/bin/activate

3 Install requirements

pip install -r requirements.txt

4 Run service

python server.py

5 Go to http://localhost:8200

Backend Config

By default service.py uses config.json. But you may call it with your own config:

python service.py -c <your_config.json>

Options

• Label_config is the most important key, it implements task logic and view for Label tool.

"label_config": "../examples/chatbot/config.xml"

• input_path for tasks: it can be a file or a directory. In case of directory all tasks will be merged together.

"input_path": "./input/tasks.json",

• output_dir is used to store completions (labeling results) in json format. output_dir will be created automatically. Each task is mapped to corresponding completion json file.

"output_dir": "./output",

Example: task a.json and b.json consist of 3 tasks and there will be 6 completion files in output_dir:

input/a.json = [{"text": "0"}, {"text": "1"}, {"text": "2"}]
input/b.json = [{"text": "3"}, {"text": "4"}, {"text": "5"}]

output/0.json = {"completions": [{"result": [...]}], "task": {"text": "0"}}  # a.json
output/1.json = {"completions": [{"result": [...]}], "task": {"text": "1"}}  # a.json
output/2.json = {"completions": [{"result": [...]}], "task": {"text": "2"}}  # a.json
output/3.json = {"completions": [{"result": [...]}], "task": {"text": "3"}}  # b.json
output/4.json = {"completions": [{"result": [...]}], "task": {"text": "4"}}  # b.json
output/5.json = {"completions": [{"result": [...]}], "task": {"text": "5"}}  # b.json

• Show instruction to person who makes labelling

"instruction": "Type something to label experts!",

• Label tool (we call it Editor) section, build_path - this points to the directory with js, css and other media from Editor (React app).

  "editor": {
    "build_path": "../build/static",
    "debug": false,
    "interfaces": [
      "basic",
      "panel",
      "submit",
      "load",
      "side-column",
      "skip",
      "controls",
      "check-empty",
      "predictions",
      "predictions:menu",
      "completions",
      "completions:menu"
    ],
  },  

• Name of your service for web

  "title": "Label Studio",

• Web server settings

  "port": 8200,
  "debug": true,

• Python logger settings are concentrated in "logger" dict.

Analytics

We collect some anonymized usage statistics that helps us better understand how Label Studio is used, and how to improve it in the future. If you want to contribute to our product, it will be highly appreciated if you are not turning off this option. However you can always do that by specifying in config:

"collect_analytics": false

Usage statistics

We collect the following statistics which we consider as completely anonymized (if you are not agree with that, please let us know):

• User ID generated by uuid4
• Label Studio version
• The mappings between labeling types and data source types, e.g. for the following config:

  <View>
      <Text name="mytext" value="$text"></Text>
      <Choices name="mytext_class" toName="mytext" choice="single">
        <Choice value="important" alias="Important document"></Choice>
        <Choice value="other" alias="Other"></Choice>
      </Choices>
  </View>

  we gather only

  [{"Choices":["Text"]}]