Skip to content

Latest commit

 

History

History
269 lines (211 loc) · 7.91 KB

README.md

File metadata and controls

269 lines (211 loc) · 7.91 KB

🐦 pigeonXT - Quickly annotate data in Jupyter Lab

PigeonXT is an extention to the original Pigeon, created by Anastasis Germanidis. PigeonXT is a simple widget that lets you quickly annotate a dataset of unlabeled examples from the comfort of your Jupyter notebook.

PigeonXT currently support the following annotation tasks:

  • binary / multi-class classification
  • multi-label classification
  • regression tasks
  • captioning tasks

Anything that can be displayed on Jupyter (text, images, audio, graphs, etc.) can be displayed by pigeon by providing the appropriate display_fn argument.

Additionally, custom hooks can be attached to each row update (example_process_fn), or when the annotating task is complete(final_process_fn).

There is a full blog post on the usage of PigeonXT on Towards Data Science.

Contributors

  • Anastasis Germanidis
  • Dennis Bakhuis
  • Ritesh Agrawal
  • Deepak Tunuguntla
  • Bram van Es

Installation

PigeonXT obviously needs a Jupyter Lab environment. Futhermore, it requires ipywidgets. The widget itself can be installed using pip:

    pip install pigeonXT-jupyter

Currently, it is much easier to install due to Jupyterlab 3: To run the provided examples in a new environment using Conda:

    conda create --name pigeon python=3.9
    conda activate pigeon
    pip install numpy pandas jupyterlab ipywidgets pigeonXT-jupyter

For an older Jupyterlab or any other trouble, please try the old method:

    conda create --name pigeon python=3.7
    conda activate pigeon
    conda install nodejs
    pip install numpy pandas jupyterlab ipywidgets
    jupyter nbextension enable --py widgetsnbextension
    jupyter labextension install @jupyter-widgets/jupyterlab-manager

    pip install pigeonXT-jupyter

Starting Jupyter Lab environment:

    jupyter lab

Development environment

I have moved the development environment to Poetry. To create an identical environment use:

conda env create -f environment.yml
conda activate pigeonxt
poetry install
pre-commit install

Examples

Examples are also provided in the accompanying notebook.

Binary or multi-class text classification

Code:

    import pandas as pd
    import pigeonXT as pixt

    annotations = pixt.annotate(
        ['I love this movie', 'I was really disappointed by the book'],
        options=['positive', 'negative', 'inbetween']
    )

Preview: Jupyter notebook multi-class classification

Multi-label text classification

Code:

    import pandas as pd
    import pigeonXT as pixt

    df = pd.DataFrame([
        {'example': 'Star wars'},
        {'example': 'The Positively True Adventures of the Alleged Texas Cheerleader-Murdering Mom'},
        {'example': 'Eternal Sunshine of the Spotless Mind'},
        {'example': 'Dr. Strangelove or: How I Learned to Stop Worrying and Love the Bomb'},
        {'example': 'Killer klowns from outer space'},
    ])

    labels = ['Adventure', 'Romance', 'Fantasy', 'Science fiction', 'Horror', 'Thriller']

    annotations = pixt.annotate(
        df,
        options=labels,
        task_type='multilabel-classification',
        buttons_in_a_row=3,
        reset_buttons_after_click=True,
        include_next=True,
        include_back=True,
    )

Preview: Jupyter notebook multi-label classification

Image classification

Code:

    import pandas as pd
    import pigeonXT as pixt

    from IPython.display import display, Image

    annotations = pixt.annotate(
      ['assets/img_example1.jpg', 'assets/img_example2.jpg'],
      options=['cat', 'dog', 'horse'],
      display_fn=lambda filename: display(Image(filename))
    )

Preview: Jupyter notebook multi-label classification

Audio classification

Code:

    import pandas as pd
    import pigeonXT as pixt

    from IPython.display import Audio

    annotations = pixt.annotate(
        ['assets/audio_1.mp3', 'assets/audio_2.mp3'],
        task_type='regression',
        options=(1,5,1),
        display_fn=lambda filename: display(Audio(filename, autoplay=True))
    )

    annotations

Preview: Jupyter notebook multi-label classification

multi-label text classification with custom hooks

Code:

    import pandas as pd
    import numpy as np

    from pathlib import Path
    from pigeonXT import annotate

    df = pd.DataFrame([
        {'example': 'Star wars'},
        {'example': 'The Positively True Adventures of the Alleged Texas Cheerleader-Murdering Mom'},
        {'example': 'Eternal Sunshine of the Spotless Mind'},
        {'example': 'Dr. Strangelove or: How I Learned to Stop Worrying and Love the Bomb'},
        {'example': 'Killer klowns from outer space'},
    ])

    labels = ['Adventure', 'Romance', 'Fantasy', 'Science fiction', 'Horror', 'Thriller']
    shortLabels = ['A', 'R', 'F', 'SF', 'H', 'T']

    df.to_csv('inputtestdata.csv', index=False)


    def setLabels(labels, numClasses):
        row = np.zeros([numClasses], dtype=np.uint8)
        row[labels] = 1
        return row

    def labelPortion(
        inputFile,
        labels = ['yes', 'no'],
        outputFile='output.csv',
        portionSize=2,
        textColumn='example',
        shortLabels=None,
    ):
        if shortLabels == None:
            shortLabels = labels

        out = Path(outputFile)
        if out.exists():
            outdf = pd.read_csv(out)
            currentId = outdf.index.max() + 1
        else:
            currentId = 0

        indf = pd.read_csv(inputFile)
        examplesInFile = len(indf)
        indf = indf.loc[currentId:currentId + portionSize - 1]
        actualPortionSize = len(indf)
        print(f'{currentId + 1} - {currentId + actualPortionSize} of {examplesInFile}')
        sentences = indf[textColumn].tolist()

        for label in shortLabels:
            indf[label] = None

        def updateRow(example, selectedLabels):
            print(example, selectedLabels)
            labs = setLabels([labels.index(y) for y in selectedLabels], len(labels))
            indf.loc[indf[textColumn] == example, shortLabels] = labs

        def finalProcessing(annotations):
            if out.exists():
                prevdata = pd.read_csv(out)
                outdata = pd.concat([prevdata, indf]).reset_index(drop=True)
            else:
                outdata = indf.copy()
            outdata.to_csv(out, index=False)

        annotated = annotate(
            sentences,
            options=labels,
            task_type='multilabel-classification',
            buttons_in_a_row=3,
            reset_buttons_after_click=True,
            include_next=False,
            example_process_fn=updateRow,
            final_process_fn=finalProcessing
        )
        return indf

    def getAnnotationsCountPerlabel(annotations, shortLabels):

        countPerLabel = pd.DataFrame(columns=shortLabels, index=['count'])

        for label in shortLabels:
            countPerLabel.loc['count', label] = len(annotations.loc[annotations[label] == 1.0])

        return countPerLabel

    def getAnnotationsCountPerlabel(annotations, shortLabels):

        countPerLabel = pd.DataFrame(columns=shortLabels, index=['count'])

        for label in shortLabels:
            countPerLabel.loc['count', label] = len(annotations.loc[annotations[label] == 1.0])

        return countPerLabel


    annotations = labelPortion('inputtestdata.csv',
                               labels=labels,
                               shortLabels= shortLabels)

    # counts per label
    getAnnotationsCountPerlabel(annotations, shortLabels)

Preview: Jupyter notebook multi-label classification

The complete and runnable examples are available in the provided Notebook.