Main Content

knnsearch

Find k-nearest neighbors using input data

Description

Idx = knnsearch(X,Y) finds the nearest neighbor in X for each query point in Y and returns the indices of the nearest neighbors in Idx, a column vector. Idx has the same number of rows as Y.

example

Idx = knnsearch(X,Y,Name,Value) returns Idx with additional options specified using one or more name-value pair arguments. For example, you can specify the number of nearest neighbors to search for and the distance metric used in the search.

example

[Idx,D] = knnsearch(___) additionally returns the matrix D, using any of the input arguments in the previous syntaxes. D contains the distances between each observation in Y and the corresponding closest observations in X.

example

Examples

collapse all

Find the patients in the hospital data set that most closely resemble the patients in Y, according to age and weight.

Load the hospital data set.

load hospital;
X = [hospital.Age hospital.Weight];
Y = [20 162; 30 169; 40 168; 50 170; 60 171];   % New patients

Perform a knnsearch between X and Y to find indices of nearest neighbors.

Idx = knnsearch(X,Y);

Find the patients in X closest in age and weight to those in Y.

X(Idx,:)
ans = 5×2

    25   171
    25   171
    39   164
    49   170
    50   172

Find the 10 nearest neighbors in X to each point in Y, first using the Minkowski distance metric and then using the Chebychev distance metric.

Load Fisher's iris data set.

load fisheriris
X = meas(:,3:4);    % Measurements of original flowers
Y = [5 1.45;6 2;2.75 .75];  % New flower data

Perform a knnsearch between X and the query points Y using Minkowski and Chebychev distance metrics.

[mIdx,mD] = knnsearch(X,Y,'K',10,'Distance','minkowski','P',5);
[cIdx,cD] = knnsearch(X,Y,'K',10,'Distance','chebychev');

Visualize the results of the two nearest neighbor searches. Plot the training data. Plot the query points with the marker X. Use circles to denote the Minkowski nearest neighbors. Use pentagrams to denote the Chebychev nearest neighbors.

gscatter(X(:,1),X(:,2),species)
line(Y(:,1),Y(:,2),'Marker','x','Color','k',...
   'Markersize',10,'Linewidth',2,'Linestyle','none')
line(X(mIdx,1),X(mIdx,2),'Color',[.5 .5 .5],'Marker','o',...
   'Linestyle','none','Markersize',10)
line(X(cIdx,1),X(cIdx,2),'Color',[.5 .5 .5],'Marker','p',...
   'Linestyle','none','Markersize',10)
legend('setosa','versicolor','virginica','query point',...
'minkowski','chebychev','Location','best')

Figure contains an axes object. The axes object contains 6 objects of type line. One or more of the lines displays its values using only markers These objects represent setosa, versicolor, virginica, query point, minkowski, chebychev.

Create two large matrices of points, and then measure the time used by knnsearch with the default "euclidean" distance metric.

rng default % For reproducibility
N = 10000;
X = randn(N,1000);
Y = randn(N,1000);
Idx = knnsearch(X,Y); % Warm up function for more reliable timing information
tic
Idx = knnsearch(X,Y);
standard = toc
standard = 
15.0068

Next, measure the time used by knnsearch with the "fasteuclidean" distance metric. Specify a cache size of 100.

Idx2 = knnsearch(X,Y,Distance="fasteuclidean",CacheSize=100); % Warm up function
tic
Idx2 = knnsearch(X,Y,Distance="fasteuclidean",CacheSize=100);
accelerated = toc
accelerated = 
1.3934

Evaluate how many times faster the accelerated computation is compared to the standard.

standard/accelerated
ans = 
10.7696

The accelerated version is more than three times faster for this example.

Input Arguments

collapse all

Input data, specified as a numeric matrix. Rows of X correspond to observations, and columns correspond to variables.

Data Types: single | double

Query points, specified as a numeric matrix. Rows of Y correspond to observations, and columns correspond to variables. Y must have the same number of columns as X.

Data Types: single | double

Name-Value Arguments

Specify optional pairs of arguments as Name1=Value1,...,NameN=ValueN, where Name is the argument name and Value is the corresponding value. Name-value arguments must appear after other arguments, but the order of the pairs does not matter.

Before R2021a, use commas to separate each name and value, and enclose Name in quotes.

Example: knnsearch(X,Y,'K',10,'IncludeTies',true,'Distance','cityblock') searches for 10 nearest neighbors, including ties and using the city block distance.

Number of nearest neighbors to find in X for each point in Y, specified as the comma-separated pair consisting of 'K' and a positive integer.

Example: 'K',10

Data Types: single | double

Flag to include all nearest neighbors that have the same distance from query points, specified as the comma-separated pair consisting of 'IncludeTies' and false (0) or true (1).

If 'IncludeTies' is false, then knnsearch chooses the observation with the smallest index among the observations that have the same distance from a query point.

If 'IncludeTies' is true, then:

  • knnsearch includes all nearest neighbors whose distances are equal to the kth smallest distance in the output arguments. To specify k, use the 'K' name-value pair argument.

  • Idx and D are m-by-1 cell arrays such that each cell contains a vector of at least k indices and distances, respectively. Each vector in D contains distances arranged in ascending order. Each row in Idx contains the indices of the nearest neighbors corresponding to the distances in D.

Example: 'IncludeTies',true

Nearest neighbor search method, specified as the comma-separated pair consisting of 'NSMethod' and one of these values.

  • 'kdtree' — Creates and uses a Kd-tree to find nearest neighbors. 'kdtree' is the default value when the number of columns in X is less than or equal to 10, X is not sparse, and the distance metric is 'euclidean', 'cityblock', 'chebychev', or 'minkowski'. Otherwise, the default value is 'exhaustive'.

    The value 'kdtree' is valid only when the distance metric is one of the four metrics noted above.

  • 'exhaustive' — Uses the exhaustive search algorithm by computing the distance values from all the points in X to each point in Y.

Example: 'NSMethod','exhaustive'

Distance metric knnsearch uses, specified as one of the values in this table or a function handle.

ValueDescription
'euclidean'Euclidean distance
'seuclidean'Standardized Euclidean distance. Each coordinate difference between the rows in X and the query matrix Y is scaled by dividing by the corresponding element of the standard deviation computed from X. To specify a different scaling, use the 'Scale' name-value argument.
'fasteuclidean'Euclidean distance computed by using an alternative algorithm that saves time when the number of predictors is at least 10. In some cases, this faster algorithm can reduce accuracy. This distance metric is available only when NSMethod is 'exhaustive'. Algorithms starting with 'fast' do not support sparse data. For details, see Algorithms.
'fastseuclidean'Standardized Euclidean distance computed by using an alternative algorithm that saves time when the number of predictors is at least 10. In some cases, this faster algorithm can reduce accuracy. This distance metric is available only when NSMethod is 'exhaustive'. Algorithms starting with 'fast' do not support sparse data. For details, see Algorithms.
'cityblock'City block distance
'chebychev'Chebychev distance (maximum coordinate difference)
'minkowski'Minkowski distance. The default exponent is 2. To specify a different exponent, use the 'P' name-value argument.
'mahalanobis'Mahalanobis distance, computed using a positive definite covariance matrix. To change the value of the covariance matrix, use the 'Cov' name-value argument.
'cosine'One minus the cosine of the included angle between observations (treated as vectors)
'correlation'One minus the sample linear correlation between observations (treated as sequences of values)
'spearman'One minus the sample Spearman's rank correlation between observations (treated as sequences of values)
'hamming'Hamming distance, which is the percentage of coordinates that differ
'jaccard'One minus the Jaccard coefficient, which is the percentage of nonzero coordinates that differ

You can also specify a function handle for a custom distance metric by using @ (for example, @distfun). A custom distance function must:

  • Have the form function D2 = distfun(ZI,ZJ).

  • Take as arguments:

    • A 1-by-n vector ZI containing a single row from X or from the query points Y.

    • An m2-by-n matrix ZJ containing multiple rows of X or Y.

  • Return an m2-by-1 vector of distances D2, whose jth element is the distance between the observations ZI and ZJ(j,:).

For more information, see Distance Metrics.

Example: 'Distance','chebychev'

Data Types: char | string | function_handle

Size of the Gram matrix in megabytes, specified as a positive scalar or "maximal". The knnsearch function can use CacheSize only when the Distance name-value argument begins with fast and the NSMethod name-value argument is set to 'exhaustive'.

If you set CacheSize to "maximal", knnsearch tries to allocate enough memory for an entire intermediate matrix whose size is MX-by-MY, where MX is the number of rows of the input data X, and MY is the number of rows of the input data Y. The cache size does not have to be large enough for an entire intermediate matrix, but must be at least large enough to hold an MX-by-1 vector. Otherwise, knnsearch uses the standard algorithm for computing Euclidean distance.

If the value of the Distance argument begins with fast, the value of NSMethod is 'exhaustive', and the value of CacheSize is too large or "maximal", knnsearch might try to allocate a Gram matrix that exceeds the available memory. In this case, MATLAB® issues an error.

Example: CacheSize="maximal"

Data Types: double | char | string

Exponent for the Minkowski distance metric, specified as the comma-separated pair consisting of 'P' and a positive scalar.

This argument is valid only if 'Distance' is 'minkowski'.

Example: 'P',3

Data Types: single | double

Covariance matrix for the Mahalanobis distance metric, specified as the comma-separated pair consisting of 'Cov' and a positive definite matrix.

This argument is valid only if 'Distance' is 'mahalanobis'.

Example: 'Cov',eye(4)

Data Types: single | double

Scale parameter value for the standardized Euclidean distance metric, specified as the comma-separated pair consisting of 'Scale' and a nonnegative numeric vector. 'Scale' has length equal to the number of columns in X. When knnsearch computes the standardized Euclidean distance, each coordinate of X is scaled by the corresponding element of 'Scale', as is each query point. This argument is valid only when 'Distance' is 'seuclidean'.

Example: 'Scale',quantile(X,0.75) - quantile(X,0.25)

Data Types: single | double

Maximum number of data points in the leaf node of the Kd-tree, specified as the comma-separated pair consisting of 'BucketSize' and a positive integer. This argument is valid only when NSMethod is 'kdtree'.

Example: 'BucketSize',20

Data Types: single | double

Flag to sort returned indices according to distance, specified as the comma-separated pair consisting of 'SortIndices' and either true (1) or false (0).

For faster performance, you can set SortIndices to false when the following are true:

  • Y contains many observations that have many nearest neighbors in X.

  • NSMethod is 'kdtree'.

  • IncludeTies is false.

In this case, knnsearch returns the indices of the nearest neighbors in no particular order. When SortIndices is true, the function arranges the nearest neighbor indices in ascending order by distance.

SortIndices is true by default. When NSMethod is 'exhaustive' or IncludeTies is true, the function always sorts the indices.

Example: 'SortIndices',false

Data Types: logical

Output Arguments

collapse all

Input data indices of the nearest neighbors, returned as a numeric matrix or cell array of numeric vectors.

  • If you do not specify IncludeTies (false by default), then Idx is an m-by-k numeric matrix, where m is the number of rows in Y and k is the number of searched nearest neighbors. Idx(j,i) indicates that X(Idx(j,i),:) is one of the k closest observations in X to the query point Y(j,:).

  • If you specify 'IncludeTies',true, then Idx is an m-by-1 cell array such that cell j (Idx{j}) contains a vector of at least k indices of the closest observations in X to the query point Y(j,:).

If SortIndices is true, then knnsearch arranges the indices in ascending order by distance.

Distances of the nearest neighbors to the query points, returned as a numeric matrix or cell array of numeric vectors.

  • If you do not specify IncludeTies (false by default), then D is an m-by-k numeric matrix, where m is the number of rows in Y and k is the number of searched nearest neighbors. D(j,i) is the distance between X(Idx(j,i),:) and Y(j,:) with respect to the distance metric.

  • If you specify 'IncludeTies',true, then D is an m-by-1 cell array such that cell j (D{j}) contains a vector of at least k distances of the closest observations in X to the query point Y(j,:).

If SortIndices is true, then knnsearch arranges the distances in ascending order.

Tips

  • For a fixed positive integer k, knnsearch finds the k points in X that are the nearest to each point in Y. To find all points in X within a fixed distance of each point in Y, use rangesearch.

  • knnsearch does not save a search object. To create a search object, use createns.

Algorithms

collapse all

For information on a specific search algorithm, see k-Nearest Neighbor Search and Radius Search.

Fast Euclidean Distance Algorithm

The values of the Distance argument that begin fast (such as 'fasteuclidean' and 'fastseuclidean') calculate Euclidean distances using an algorithm that uses extra memory to save computational time. This algorithm is named "Euclidean Distance Matrix Trick" in Albanie [1] and elsewhere. Internal testing shows that this algorithm saves time when the number of predictors is at least 10. Algorithms starting with 'fast' do not support sparse data.

To find the matrix D of distances between all the points xi and xj, where each xi has n variables, the algorithm computes distance using the final line in the following equations:

Di,j2=xixj2=(xixj)T(xixj)=xi22xiTxj+xj2.

The matrix xiTxj in the last line of the equations is called the Gram matrix. Computing the set of squared distances is faster, but slightly less numerically stable, when you compute and use the Gram matrix instead of computing the squared distances by squaring and summing. For a discussion, see Albanie [1].

To store the Gram matrix, the software uses a cache with the default size of 1e3 megabytes. You can set the cache size using the CacheSize name-value argument. If the value of CacheSize is too large or "maximal", knnsearch might try to allocate a Gram matrix that exceeds the available memory. In this case, MATLAB issues an error.

References

[1] Albanie, Samuel. Euclidean Distance Matrix Trick. June, 2019. Available at https://www.robots.ox.ac.uk/%7Ealbanie/notes/Euclidean_distance_trick.pdf.

Alternative Functionality

If you set the knnsearch function's 'NSMethod' name-value pair argument to the appropriate value ('exhaustive' for an exhaustive search algorithm or 'kdtree' for a Kd-tree algorithm), then the search results are equivalent to the results obtained by conducting a distance search using the knnsearch object function. Unlike the knnsearch function, the knnsearch object function requires an ExhaustiveSearcher or a KDTreeSearcher model object.

Simulink Block

To integrate a k-nearest neighbor search into Simulink®, you can use the KNN Search block in the Statistics and Machine Learning Toolbox™ library or a MATLAB Function block with the knnsearch function. For an example, see Predict Class Labels Using MATLAB Function Block.

When deciding which approach to use, consider the following:

  • If you use the Statistics and Machine Learning Toolbox library block, you can use the Fixed-Point Tool (Fixed-Point Designer) to convert a floating-point model to fixed point.

  • Support for variable-size arrays must be enabled for a MATLAB Function block with the knnsearch function.

References

[1] Friedman, J. H., J. Bentley, and R. A. Finkel. “An Algorithm for Finding Best Matches in Logarithmic Expected Time.” ACM Transactions on Mathematical Software 3, no. 3 (1977): 209–226.

Extended Capabilities

Version History

Introduced in R2010a

expand all