swarmauri_measurement_mutualinformation 0.9.0

Swarmauri Mutual Information Measurement Community Package.

---

Swarmauri Measurement Mutual Information

Mutual-information measurement plugin for Swarmauri pipelines. Computes the average mutual information (in bits) between every feature column and a target column, letting you rank signal strength before training models.

Features

• Wraps sklearn.feature_selection.mutual_info_classif behind the standard MeasurementBase API.
• Supports Pandas DataFrame inputs; automatically excludes the target column from the feature set.
• Returns the average mutual information across all features (in bits) for quick screening.

Prerequisites

• Python 3.10 or newer.
• scikit-learn and pandas installed (pulled in as dependencies of this package).
• Clean, pre-processed categorical data (encode non-numeric columns before calling) since mutual_info_classif expects numerical inputs.

Installation

# pip
pip install swarmauri_measurement_mutualinformation

# poetry
poetry add swarmauri_measurement_mutualinformation

# uv (pyproject-based projects)
uv add swarmauri_measurement_mutualinformation

Quickstart

import pandas as pd
from swarmauri_measurement_mutualinformation import MutualInformationMeasurement

# Example dataset
frame = pd.DataFrame(
    {
        "feature_a": [0, 1, 1, 0, 1, 0],
        "feature_b": [5.1, 5.0, 4.9, 5.2, 5.1, 5.0],
        "target": [0, 1, 1, 0, 1, 0],
    }
)

mi = MutualInformationMeasurement()
avg_mi = mi.calculate(frame, target_column="target")
print(f"Average mutual information: {avg_mi:.4f} bits")

Per-Feature Scores

If you need the individual MI score per feature, compute it directly and inspect the array:

import pandas as pd
from sklearn.feature_selection import mutual_info_classif

frame = pd.DataFrame(
    {
        "feat1": [0, 1, 1, 0, 1, 0],
        "feat2": [5.1, 5.0, 4.9, 5.2, 5.1, 5.0],
        "target": [0, 1, 1, 0, 1, 0],
    }
)

scores = mutual_info_classif(frame[["feat1", "feat2"]], frame["target"])
for column, score in zip(["feat1", "feat2"], scores):
    print(column, score)

Use the per-feature scores to filter low-signal columns before passing the DataFrame back through Swarmauri.

Tips

• Normalize or discretize continuous features when comparing very different scales; mutual information is sensitive to distribution assumptions.
• Handle missing values before calling calculate; mutual_info_classif does not accept NaNs.
• Binary targets work out of the box; for multi-class targets, ensure target_column contains integer encodings.

Want to help?

If you want to contribute to swarmauri-sdk, read up on our guidelines for contributing that will help you get started.