emlangkit 0.0.1

Emergent Language Analysis Toolkit

Emergent Language Analysis Toolkit

This toolkit aims to collect all metrics currently used in emergent communication research into one place. The usage should be convenient and the inputs should be standardised, to ease adoption and spread of these metrics.

Installation

To install emlangkit, run pip install emlangkit.

Automatic tests are run for Python 3.9, 3.10, 3.11.

Usage

All metrics are available through the Language class in emlangkit.Language. This class accepts two numpy arrays as inputs - messages and observations. These are then used, with some possible speedups, to calculate any requested metric, as per example below

import numpy as np
from emlangkit import Language

messages = np.array(
    [[1, 2, 0, 3, 4], [1, 2, 2, 3, 4], [1, 2, 2, 3, 0], [1, 0, 0, 1, 2]]
)
observations = np.array([[4, 4], [4, 3], [3, 2], [1, 4]])

lang = Language(messages=messages, observations=observations)

score, p_value = lang.topsim()

# Mutual information already requires both language and observation entropy
mi = lang.mutual_information()

# So this call uses less computation
lang_entropy = lang.language_entropy()

Metrics

Currently available metrics, with their implementations as per below.

• Entropy (emlangkit.metrics.entropy)
• Mutual Information (emlangkit.metrics.mutual_information)
• Topographic Similarity (emlangkit.metrics.topsim)
• Positional Disentanglement (emlangkit.metrics.posdis)
• Bag-of-Words Disentanglement (emlangkit.metrics.bosdis)

Contributing

All pull requests are welcome! Just please make sure to install pre-commit and run the pytests before submitting a PR. Additionally, if a lot of new code is added, please also add the relevant tests.

Related Libraries

This is a non-exhaustive list of libraries related to EC research. Please feel free to open a PR to add to it!

• EGG - https://github.com/facebookresearch/EGG
• Harris' Articulation Scheme - https://github.com/wedddy0707/HarrisSegmentation
• CGI - https://github.com/wedddy0707/categorial_grammar_induction_of_emergent_language

Sources

Most of the base metrics are inspired or taken from either EGG, or code from the paper "Catalytic Role Of Noise And Necessity Of Inductive Biases In The Emergence Of Compositional Communication" here.

Citation

If you find emlangkit useful in your work, please cite it as below:

@software{lipinski_emlangkit_2023,
        title = {emlangkit: Emergent Language Analysis Toolkit},
        url = {https://github.com/olipinski/emlangkit},
        author = {Lipinski, Olaf},
        year = {2023}
}