Mean Shift

Algorithm overview

MeanShift detects high-density regions and infers cluster count automatically from data density.

This algorithm is especially useful when:

• You want clustering without specifying the number of clusters.
• Density peaks are more meaningful than centroid partitions.
• Your product needs adaptive grouping behavior over time.

JavaScript implementation

@kanaries/ml gives JavaScript applications access to Mean Shift for cases where clusters form around dense modes rather than around a fixed, user-specified cluster count. That can be useful in browser-based exploratory tools where users want clustering without deciding k up front.

Because the implementation is available in JS, you can experiment with bandwidth-driven clustering directly inside Node.js services or interactive frontend tools.

Quick start

from sklearn.cluster import MeanShift

X = [[0, 0], [0.1, 0.2], [3, 3], [3.2, 3.1]]

model = MeanShift(bandwidth=1.0)
labels = model.fit_predict(X)

import { Clusters } from '@kanaries/ml';

const X = [[0, 0], [0.1, 0.2], [3, 3], [3.2, 3.1]];

const model = new Clusters.MeanShift(1.0);
const labels = model.fitPredict(X);

Quick JavaScript example

import { Clusters } from '@kanaries/ml';

const X = [[0, 0], [0.1, 0.2], [3, 3], [3.2, 3.1]];

const model = new Clusters.MeanShift(1.0);
const labels = model.fitPredict(X);

Detailed API reference

constructor(
    bandwidth: number = 1,
    max_iter: number = 300,
    distanceType: Distance.IDistanceType = 'euclidean'
)

Methods:

• fitPredict(samplesX: number[][]): number[]
• getCentroids(): number[][]

const ms = new Clusters.MeanShift(2);
const labels = ms.fitPredict(X);
const centers = ms.getCentroids();

Implementation workflow

1. Scale numeric features and choose bandwidth heuristics.
2. Fit and inspect discovered modes and assigned labels.
3. Tune bandwidth to balance over-fragmentation and over-merging.

JavaScript deployment notes

• Spend time tuning bandwidth because it strongly controls cluster granularity.
• Mean Shift is better suited to exploratory analysis than very large real-time workloads.
• For browser apps, prefer smaller datasets or background workers because iterative shifting can be expensive.