探索モデル オブジェクトを使用して k 最近傍を探索 - MATLAB knnsearch - MathWorks 日本 (2024)

ライブ スクリプトを開く

load fisheririsrng(1); % For reproducibilityn = size(meas,1);idx = randsample(n,5);X = meas(~ismember(1:n,idx),:); % Training dataY = meas(idx,:); % Query data

MdlKDT = KDTreeSearcher(X)

MdlKDT = KDTreeSearcher with properties: BucketSize: 50 Distance: 'euclidean' DistParameter: [] X: [145x4 double]

MdlES = ExhaustiveSearcher(X)

MdlES = ExhaustiveSearcher with properties: Distance: 'euclidean' DistParameter: [] X: [145x4 double]

IdxKDT = knnsearch(MdlKDT,Y);IdxES = knnsearch(MdlES,Y);[IdxKDT IdxES]

ans = 5×2 17 17 6 6 1 1 89 89 124 124

ライブ スクリプトを開く

load fisheriris

rng(1) % For reproducibilityn = size(meas,1); % Sample sizeqIdx = randsample(n,5); % Indices of query datatIdx = ~ismember(1:n,qIdx); % Indices of training dataQ = meas(qIdx,:);X = meas(tIdx,:);

Mdl = createns(X,'Distance','minkowski')

Mdl = KDTreeSearcher with properties: BucketSize: 50 Distance: 'minkowski' DistParameter: 2 X: [145x4 double]

IdxNN = knnsearch(Mdl,Q,'K',2)

IdxNN = 5×2 17 4 6 2 1 12 89 66 124 100

ライブ スクリプトを開く

load fisheriris

rng(4); % For reproducibilityn = size(meas,1); % Sample sizeqIdx = randsample(n,5); % Indices of query dataX = meas(~ismember(1:n,qIdx),:);Y = meas(qIdx,:);

Mdl = KDTreeSearcher(X);

[Idx,D] = knnsearch(Mdl,Y,'K',7,'IncludeTies',true);

cellfun('length',Idx)

ans = 5×1 8 7 7 7 7

nn5 = Idx{1}

nn5 = 1×8 91 98 67 69 71 93 88 95

nn5d = D{1}

nn5d = 1×8 0.1414 0.2646 0.2828 0.3000 0.3464 0.3742 0.3873 0.3873

ライブ スクリプトを開く

load fisheririsX = meas(:,3:4); % PredictorsY = species; % Response

KDTreeMdl = KDTreeSearcher(X,'Distance','minkowski','P',5)

KDTreeMdl = KDTreeSearcher with properties: BucketSize: 50 Distance: 'minkowski' DistParameter: 5 X: [150x2 double]

newpoint = [5 1.45];[IdxMk,DMk] = knnsearch(KDTreeMdl,newpoint,'k',10);[IdxCb,DCb] = knnsearch(KDTreeMdl,newpoint,'k',10,'Distance','chebychev');

figure;gscatter(X(:,1),X(:,2),Y);title('Fisher''s Iris Data -- Nearest Neighbors');xlabel('Petal length (cm)');ylabel('Petal width (cm)');hold onplot(newpoint(1),newpoint(2),'kx','MarkerSize',10,'LineWidth',2); % Query point plot(X(IdxMk,1),X(IdxMk,2),'o','Color',[.5 .5 .5],'MarkerSize',10); % Minkowski nearest neighborsplot(X(IdxCb,1),X(IdxCb,2),'p','Color',[.5 .5 .5],'MarkerSize',10); % Chebychev nearest neighborslegend('setosa','versicolor','virginica','query point',... 'minkowski','chebychev','Location','Best');

h = gca; % Get current axis handle.h.XLim = [4.5 5.5];h.YLim = [1 2];axis square;

ライブ スクリプトを開く

rng default % For reproducibilityA = diag(1:100);B = randn(100,10);K = kron(A,B);ss = size(K)

ss = 1×2 10000 1000

tic;h = hnswSearcher(K);chnsw = toc

chnsw = 10.6544

Y = randn(1000, 1000);tic;[idx, D] = knnsearch(h,Y,K=5);thnsw = toc

thnsw = 0.0797

tice = ExhaustiveSearcher(K);ces = toc

ces = 0.0021

tic;[idx0, D0] = knnsearch(e,Y,K=5);tes = toc

tes = 1.4841

v = abs(idx - idx0); % Count any difference in the five found entriesvv = sum(v,2); % vv = 0 means no differencenz = nnz(vv) % Out of 1000 rows how many differ at all?

nz = 118

tic;h2 = hnswSearcher(K,MaxNumLinksPerNode=48,TrainSetSize=2000);chnsw2 = toc

chnsw2 = 78.4836

tic;[idx2, D2] = knnsearch(h2,Y,K=5);thnsw2 = toc

thnsw2 = 0.1049

v = abs(idx2 - idx0);vv = sum(v,2);nz = nnz(vv)

nz = 57

timet = table([chnsw;ces;chnsw2],[thnsw;tes;thnsw2],'RowNames',["HNSW";"Exhaustive";"HNSW2"],'VariableNames',["Creation" "Search"])

timet=3×2 table Creation Search _________ ________ HNSW 10.654 0.079741 Exhaustive 0.0021304 1.4841 HNSW2 78.484 0.10492