CoreSource checked

Perceptron

Learns a linear binary classifier by updating weights after mistakes.

Explore relationships Open constellationBrowse atlas

Quick Facts

Compact metadata for implementation and discovery.

Category: Machine Learning
Difficulty: Beginner
Family: Linear Classifier
Tier: Core
Status: Verified
Review: Source Checked
Sources: 1
Formulas: 1
Relationships: 3
Code example: Ready
Visualization: Ready

Overview

The Perceptron predicts from a linear score and adjusts weights toward misclassified positive examples and away from misclassified negative examples.

Core Concepts

linear scoremistake-driven updatedecision boundaryonline learning

Quick facts

Complexity

Runtime and memory behavior at a glance.

Best

O(d)

Average

O(e n d)

Worst

O(e n d)

Space

O(d)

When to Use

Use when a simple online linear classifier is enough.
Use when introductory classification is a central requirement.

When Not to Use

Avoid when classes are not close to linearly separable and nonlinear features are unavailable.
Avoid when the data size, labels, or feature assumptions do not match the model.

What It Does

Computes a linear score
Predicts a binary label
Updates weights on mistakes

Use Cases

Introductory classification
Online linear learning
Linearly separable data

Inputs

Feature vectors
Binary labels
Learning rate

Outputs

Linear classifier weights

Key Formulas (1)

Perceptron update

weights_next = weights + learning_rate · label · features

weights_next = weights + learning_rate * label * features

Updates weights when the current example is misclassified.

Pseudocode

1.Initialize weights to zero or small values
2.For each training example, compute the sign of the score
3.If the prediction is wrong, update weights toward the correct label
4.Repeat for several passes

Canonical Example

Input: point = [2, 1], label = +1, prediction = -1
Output: weights increase toward [2, 1]
Explanation: A mistake causes the classifier to move its boundary toward the correct side.

Implementation Notes

Shuffle examples between epochs to reduce ordering effects.
Split data for validation and keep preprocessing identical between training and inference.
Keep a tiny example with known output beside the implementation.
Treat examples as educational sketches; production models need validation, monitoring, and domain checks.

Common Pitfalls

Using labels 0 and 1 directly can break the usual update rule, which expects signs such as -1 and +1.
Evaluating only on training data can hide overfitting and poor generalization.
Not defining boundary, empty, or degenerate inputs before implementation.
Data leakage, unscaled features, and validation on training data can make results look better than they are.

Interactive Mechanism Walkthrough

Iterative Convergence & State Trajectory

Show a separating line shifting after each misclassified point.

Step 1 of 4: Initialize weights to zero or small values

model-training

Notes

Perceptron operates in O(e n d) time and O(d) space within the Linear Classifier family. Key governing relation: weights_next = weights + learning_rate * label * features. Often compared with Logistic Regression and Multilayer Perceptron.