Installation

pip install -U quica

Get Quick Agreement

If you already have a python dataframe you can run Quica with few liens of code! Let’s assume you have two coders; we will create a pandas dataframe just to show how to use the library. As for now, we support only integer values and we still have not included weighting.

from quica.quica import Quica
import pandas as pd

coder_1 = [0, 1, 0, 1, 0, 1]
coder_3 = [0, 1, 0, 1, 0, 0]

dataframe = pd.DataFrame({"coder1" : coder_1,
              "coder3" : coder_3})

quica = Quica(dataframe=dataframe)
print(quica.get_results())

This is the expected output:

Out[1]:
             score
names
krippendorff  0.685714
fleiss        0.666667
scotts        0.657143
raw           0.833333
mace          0.426531
cohen         0.666667

It was pretty easy to get all the scores, right? What if we do not have a pandas dataframe? what if we want to directly get the latex table to put into the paper? worry not, my friend: it’s easier done than said!

from quica.measures.irr import *
from quica.dataset.dataset import IRRDataset
from quica.quica import Quica

coder_1 = [0, 1, 0, 1, 0, 1]
coder_3 = [0, 1, 0, 1, 0, 0]

disagreeing_coders = [coder_1, coder_3]
disagreeing_dataset = IRRDataset(disagreeing_coders)

quica = Quica(disagreeing_dataset)

print(quica.get_results())
print(quica.get_latex())

you should get this in output, note that the latex table requires the booktabs package:

Out[1]:
             score
names
krippendorff  0.685714
fleiss        0.666667
scotts        0.657143
raw           0.833333
mace          0.426531
cohen         0.666667

Out[2]:

\begin{tabular}{lr}
\toprule
{} &     score \\
names        &           \\
\midrule
krippendorff &  0.685714 \\
fleiss       &  0.666667 \\
scotts       &  0.657143 \\
raw          &  0.833333 \\
mace         &  0.426531 \\
cohen        &  0.666667 \\
\bottomrule
\end{tabular}

Features

from quica.measures.irr import *
from quica.dataset.dataset import IRRDataset
from quica.quica import Quica

coder_1 = [0, 1, 0, 1, 0, 1]
coder_2 = [0, 1, 0, 1, 0, 1]
coder_3 = [0, 1, 0, 1, 0, 0]

agreeing_coders = [coder_1, coder_2]
agreeing_dataset = IRRDataset(agreeing_coders)

disagreeing_coders = [coder_1, coder_3]
disagreeing_dataset = IRRDataset(disagreeing_coders)

kri = Krippendorff()
cohen = CohensK()

assert kri.compute_irr(agreeing_dataset) == 1
assert kri.compute_irr(agreeing_dataset) == 1
assert cohen.compute_irr(disagreeing_dataset) < 1
assert cohen.compute_irr(disagreeing_dataset) < 1

Supported Algorithms

• MACE (Multi-Annotator Competence Estimation)

  • Hovy, D., Berg-Kirkpatrick, T., Vaswani, A., & Hovy, E. (2013, June). Learning whom to trust with MACE. In Proceedings of the 2013 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (pp. 1120-1130).

  • We define the inter coder agreeement as the average competence of the users.

• Krippendorff’s Alpha

• Cohens’ K

• Fleiss’ K

• Scotts’ PI

• Raw Agreement: Standard Accuracy

Credits

This package was created with Cookiecutter and the audreyr/cookiecutter-pypackage project template. Thanks to Pietro Lesci and Dirk Hovy for their implementation of MACE.

0.1.0 (2020-11-08)

• New API to get the output

• Fixed test cases

• Extended documentation on the README file

0.1.0 (2020-11-05)

• First release on PyPI.