DataFrameLabeler

A small ipywidget tool for labeling data frames inside jupyter.

Installation

Currently, the only way to use the DataFrameLabeler is to clone this repositroy.

Why?

This small tool was inspired by the fast.ai image cleaner widget https://docs.fast.ai/widgets.image_cleaner.html . However, I needed a tool for tabular data.

How to use?

# imports
import pandas as pd
import numpy as np
import matplotlib.pyplot as plt

from DataFrameLabeler import DataFrameLabeler

# If you have a pandas data frame where you want to assign each row eihter 'SUCCESS' or 'FAILURE'.
# like the following one.
length = 100
cols = ['A', 'B', 'C', 'D', 'E']
df = pd.DataFrame(np.random.rand(length, len(cols)), columns=cols)

# First you need a function responsible to print a single row.
def plotter(idx, row):
    fig = plt.figure()
    plt.plot([i for i in row[cols]])
    # plot should not be shown when called.
    plt.close(fig)
    return fig
    
# Afterwards, just construct a DataFrameLabeler object.
# If `target_col` exists in the data frame, its content will be used as preselection.
lbl = DataFrameLabeler(df,
                       labels=['FAILURE', 'SUCCESS'], # choices for the labels
                       plotter=plotter,               # function which plots each row
                       target_col='class_name',       # column name where the labels will be stored
                       width=3,                       # number of figures in each row
                       height=2                       # number of rows shown at once
                       )

# To obtain the newly labeled data frame call lbl.get_labeled_data()

TODO:

• rework how user defined plotter works, atm its horrifying, especially when using matplotlib
• proper styling of buttons
• allow groupby argument
• allow multi selection
• add automatic saving of intermediate result to csv or pickle file
• rethink interface
• add more unit tests
• Documentation