polyview 0.0.1

A Python library for multi-view learning: fusion, clustering and embedding.

polyview

Multi-view clustering and embedding toolkit for Python.

polyview provides sklearn-compatible algorithms for datasets where each sample is described by multiple independent feature sets (views) — such as audio + video, text + images, or multi-sensor readings.

Actively maintained successor to the dormant mvlearn library.

---

Installation

pip install polyview

For development:

git clone https://github.com/gwendal-debaussart/polyview
cd polyview
pip install -e ".[dev]"

---

Quick start

import numpy as np
import polyview as pv

# Two views of the same 100 samples
X_audio  = np.random.rand(100, 13)
X_visual = np.random.rand(100, 512)

# Wrap in a dataset container
mvd = pv.MultiViewDataset(
    [X_audio, X_visual],
    view_names=["audio", "visual"], # optional name for user readability
)

# Co-regularized multi-view K-Means
model = pv.cluster.MultiviewKMeans(n_clusters=4, random_state=0)
labels = model.fit_predict(mvd)

fused = pv.fusion.NormalizedFusion().fit_transform(mvd)

---

What's included

Module	Contents
polyview.datasets	MultiViewDataset — typed container with save/load, split, subsetting
polyview.fusion	ConcatFusion, WeightedFusion, NormalizedFusion
polyview.cluster	MultiviewKMeans, MultiviewCoRegSpectralClustering, MultiviewCoTrainSpectralClustering
polyview.embed	(coming) MVMDS, MultiviewTSNE, GCCA, ...
polyview.metrics	(coming) consensus score, view agreement
polyview.semisupervised	(coming) CoTraining, label propagation
polyview.viz	(coming) view scatter grids, embedding plots

All estimators are sklearn-compatible: they work with GridSearchCV, cross_val_score, Pipeline, and clone.

---

⚖️ License

MIT License. See LICENSE for details.