k-means
Partitioning by centroids
K-means is a method for grouping data points into clusters based on their similarity.
The algorithm assigns each point to whichever cluster has the nearest center, then recalculates centers based on the new assignments, repeating until stable.
The k refers to the number of clusters, chosen in advance: We ran k-means with k=5 to segment customers into five groups.
A workhorse of machine learning and data analysis.
Forming concentrated groups
- centroidMean position of figure
- iterationRepeated procedural pass
- convergenceProcess of approaching sameness
- assignmentAllocated work or duty
Lloyd · partitioning · initialization · Euclidean · distance · optimization · unsupervised · classification
Division into sections
- customerBuyer of goods
- imageVisual or mental likeness
- compressionForcing into smaller space
- anomalyDeparture from the expected
detection · market · research · gene · expression · document · clustering · color
Grouped collections
- centroidsGeometric mean positions
- iterationsSuccessive passes through process
- toleranceWillingness to accept difference
- initializationPreparing system for operation
seeds · restarts · threshold · hyperparameter · tuning · validation · selection · cross-validation
Globe-like in form
- assumptionUnverified belief accepted as true
- outliersExtreme data points
- sensitivityResponsiveness to stimuli
- scalingClimbing or resizing
initialization · dependence · curse · imbalanced · sizes · density · varying · convergence
clustering
Forming concentrated groups
- centroidMean position of figure
- iterationRepeated procedural pass
- convergenceProcess of approaching sameness
Lloyd · assignment · partitioning · initialization · Euclidean · distance · optimization · unsupervised · classification · segmentation
segmentation
Division into sections
- customerBuyer of goods
- imageVisual or mental likeness
- compressionForcing into smaller space
anomaly · detection · market · research · gene · expression · document · clustering · color · quantization
clusters
Grouped collections
- centroidsGeometric mean positions
- iterationsSuccessive passes through process
- toleranceWillingness to accept difference
initialization · seeds · restarts · threshold · hyperparameter · tuning · validation · selection · cross-validation
spherical
Globe-like in form
- assumptionUnverified belief accepted as true
- outliersExtreme data points
- sensitivityResponsiveness to stimuli
scaling · initialization · dependence · curse · imbalanced · sizes · density · varying · convergence · trapped
- machine learningneural networks, deep learning, backpropagation, underfitting, CNNs
- clusteringclustering, clusters, dendrogram, centroid, partitioning
- statisticsmedian, mode, variance, standard deviation, correlation
- unsupervisedunattended, unwatched, unmonitored, unguided, unchaperoned
- Statistical MethodsANOVA, confidence intervals, correlation, covariance, hypothesis testing
- dimensionality reductionPCA, eigenvalues, eigenvector, loading, component