基本ガイド
Distance Metrics in Machine Learning
Distance metrics are mathematical functions that quantify how similar or different two data points are, forming the basis for algorithms like k-nearest neighbors and clustering.
このページでは3 分で読めます
概要
Choosing the right metric matters because different metrics assume different notions of similarity, and the wrong choice can make an otherwise sound algorithm perform poorly.
ディープダイブ
A distance metric turns raw feature values into a single number representing how far apart two points are, which many algorithms rely on directly. Euclidean distance, the straight-line distance familiar from geometry, is a common choice in many introductory examples and works well when features are continuous, roughly on comparable scales, and the notion of similarity matches physical closeness. Manhattan distance, also called taxicab distance, sums the absolute differences along each dimension rather than taking the square root of squared differences; it suits grid-like movement constraints and is less sensitive to outliers in individual dimensions than Euclidean distance. Cosine distance measures orientation between nonzero vectors rather than their magnitudes, making it useful in some text and recommendation systems where a document or user profile's overall direction, such as topic balance, matters more than its raw length or intensity. Mahalanobis distance generalizes Euclidean distance by accounting for the correlations and differing variances between features, effectively normalizing the space so that features are compared fairly regardless of their original scale, which is useful in anomaly detection and multivariate outlier analysis. Hamming distance counts the number of positions at which two equal-length strings or categorical vectors differ, making it suited for categorical data, error-correcting codes, and genetic sequence comparison rather than continuous numeric data. A common misconception is that Euclidean distance is always the safe default; in high-dimensional spaces, all pairwise Euclidean distances tend to become similar, a phenomenon sometimes called the curse of dimensionality, which can make nearest-neighbor comparisons less informative; checking distance distributions and task performance can reveal when that matters.
戦略的影響
より明確な判決
これは、明確な技術的主張とマーケティング言語を区別するのに役立ちます。
費用と予算
お金や時間を費やす前に、実装に関するより良い質問をすることができます。
チームとワークフロー
共通の理解を持ったチームは、製品、ポリシー、学習に関する意思決定をより適切に行うことができます。
The Future of Distance Metrics in Machine Learning
Distance metric choice remains a foundational, largely stable part of machine learning practice, though learned distance metrics, sometimes called metric learning, continue to gain traction for specialized applications like face verification, where a neural network learns an embedding space in which a simple distance, often cosine or Euclidean, becomes meaningful after training. Expect distance metrics to remain relevant even as deep learning grows, since most embedding-based systems still rely on a classical distance function applied to learned representations rather than replacing the concept of distance entirely.
現実世界の実装
A k-nearest neighbors model predicting house prices uses Euclidean distance across square footage, number of bedrooms, and age, treating all numeric differences as straight-line distance in feature space.
A city-grid delivery routing tool uses Manhattan distance instead of Euclidean distance, since vehicles must travel along street grids rather than in straight lines, matching the metric to the real movement constraint.
A recommendation system comparing user preference vectors uses cosine distance rather than Euclidean distance, since it cares about the direction of preference patterns, such as genre balance, rather than the raw magnitude of ratings.
A DNA sequence comparison tool uses Hamming distance to count the number of positions where two equal-length genetic sequences differ, since the data is categorical rather than continuous.
リスクとガードレール
チームが異なれば、同じ用語の使用方法も異なる可能性があるため、範囲を早めに定義してください。
ベンチマークは好調に見えても、実際のパフォーマンスにはばらつきがある場合があります。
データの品質と評価計画を無視すると、多くの場合、脆弱な結果が生じます。
実装ロードマップ
必要な結果を平易な言葉で定義することから始めます。
テストする前に、成功指標と失敗条件を 1 つ選択します。
洗練されたデモセットではなく、代表的なデータを使用して小規模なパイロットを実行します。
Document where Distance Metrics in Machine Learning helps and where simpler methods are better.
探検を続けましょう
Free newsletter
Get the daily AI briefing
Three verified AI stories every weekday morning, written in plain English. Free forever, no ads.
One email each weekday. Unsubscribe in one click. We never sell or share your address.
Test yourself
Take the Distance Metrics in Machine Learning quiz
Instant feedback on every answer, and a shareable certificate with a verifiable ID once you pass a course.
Support free AI education. AI Understanding is a 501(c)(3) nonprofit — no ads, no paywall, ever. Make a donation
よくある質問
What is Distance Metrics in Machine Learning?
Distance metrics are mathematical functions that quantify how similar or different two data points are, forming the basis for algorithms like k-nearest neighbors and clustering. Choosing the right metric matters because different metrics assume different notions of similarity, and the wrong choice can make an otherwise sound algorithm perform poorly.
Why did the delivery routing example choose Manhattan distance over Euclidean distance?
Manhattan distance sums differences along grid axes, matching how vehicles actually move along city streets rather than straight-line paths.
What does cosine distance measure, as defined in the guide, that makes it suited to recommendation systems?
Cosine distance captures direction, such as genre balance in a preference vector, rather than raw magnitude, which fits recommendation use cases.
What additional information does Mahalanobis distance incorporate that Euclidean distance does not?
Mahalanobis distance uses the inverse of the feature covariance matrix, accounting for correlation and scale differences that plain Euclidean distance ignores.
For what type of data is Hamming distance specifically suited, according to the guide?
The guide describes Hamming distance as counting differing positions in equal-length strings or categorical vectors, fitting genetic sequence comparison.
What preprocessing step does the guide say is typically required before computing Euclidean distance across features?
The technical section notes that features must usually be scaled, typically via standardization, so a larger-range feature doesn't dominate the distance calculation.
学び続ける
関連ガイド
このトピックのために選ばれたその他のガイド