Krippendorff's Alpha is a commonly used inter-annotator reliability metric, but it's hard to calculate on a Dataframe. This package makes it easy.
Made with ❤️ by LightTag - The Text Annotation Tool For Teams. We use this in production to give our customers a single number to understand the quality of their labeled data. Read the blog post here
Calculating Krippendorff's Alpha assumes data is formatted in a way that just doesn't appear in the wild. We wanted a package that could read a Dataframe in the formats we see in real life and give us the Alpha in one line.
pip install simpledorff
import simpledorff
import pandas as pd
Data = pd.read_csv('./examples/from_paper.csv') #Load Your Dataframe
Data.head()
Unnamed: 0 | document_id | annotator_id | annotation | |
---|---|---|---|---|
0 | 0 | 1 | A | 1.0 |
1 | 1 | 1 | B | 1.0 |
2 | 2 | 1 | D | 1.0 |
3 | 3 | 1 | C | NaN |
4 | 4 | 2 | A | 2.0 |
simpledorff.calculate_krippendorffs_alpha_for_df(Data,experiment_col='document_id',
annotator_col='annotator_id',
class_col='annotation')
0.743421052631579