이번역페이지는최신내용을담고있지않습니다。최신내용을영문으로보려면여기를클릭하십시오。gydF4y2Ba
서포트벡터머신또는다른분류기에대해다중클래스모델피팅하기gydF4y2Ba
은테이블gydF4y2BaMdlgydF4y2Ba
= fitcecoc (gydF4y2Ba资源描述gydF4y2Ba
,gydF4y2BaResponseVarNamegydF4y2Ba
)gydF4y2Ba资源描述gydF4y2Ba
에 포함된 예측 변수와gydF4y2Ba资源描述。ResponseVarNamegydF4y2Ba
에포함된클래스레이블을사용하여전체훈련된다중클래스gydF4y2Ba오류수정출력코드(ECOC)모델gydF4y2Ba을반환합니다。gydF4y2BafitcecocgydF4y2Ba
는일대일gydF4y2Ba코딩설계gydF4y2Ba를활용하여K (K - 1) / 2개의이진서포트벡터머신(SVM)모델을사용합니다。여기K서는고유한클래스레이블(레벨)개수입니다。gydF4y2BaMdlgydF4y2Ba
은gydF4y2BaClassificationECOCgydF4y2Ba
모델입니다。gydF4y2Ba
는테이블gydF4y2BaMdlgydF4y2Ba
= fitcecoc (gydF4y2Ba资源描述gydF4y2Ba
,gydF4y2Ba公式gydF4y2Ba
)gydF4y2Ba资源描述gydF4y2Ba
에포함된예측변수와클래스레이블을사용하여ECOC모델을반환합니다。gydF4y2Ba公式gydF4y2Ba
는훈련에사용된gydF4y2Ba资源描述gydF4y2Ba
에 포함된 응답 변수와 예측 변수의 부분 집합에 대한 설명 모델입니다.gydF4y2Ba
는테이블gydF4y2BaMdlgydF4y2Ba
= fitcecoc (gydF4y2Ba资源描述gydF4y2Ba
,gydF4y2BaYgydF4y2Ba
)gydF4y2Ba资源描述gydF4y2Ba
에포함된예측변수와벡터gydF4y2BaYgydF4y2Ba
에포함된클래스레이블을사용하여ECOC모델을반환합니다。gydF4y2Ba
는예측변수gydF4y2BaMdlgydF4y2Ba
= fitcecoc (gydF4y2BaXgydF4y2Ba
,gydF4y2BaYgydF4y2Ba
)gydF4y2BaXgydF4y2Ba
와클래스레이블gydF4y2BaYgydF4y2Ba
를 사용하여 훈련된 经济合作모델을 반환합니다.gydF4y2Ba
는위에열거된구문을사용하여하나이상의gydF4y2BaMdlgydF4y2Ba
= fitcecoc (gydF4y2Ba___gydF4y2Ba,gydF4y2Ba名称,值gydF4y2Ba
)gydF4y2Ba名称,值gydF4y2Ba
쌍의인수로지정된추가옵션을통해ECOC모델을반환합니다。gydF4y2Ba
예를들어,다른이진학습기또는다른코딩설계를지정하거나교차검증할수있습니다。gydF4y2BaKfoldgydF4y2Ba
名称,值gydF4y2Ba
쌍의인수를사용하여교차검증을수행하는것이좋습니다。교차검증결과를통해모델이얼마나잘일반화되는지확인할수있습니다。gydF4y2Ba
[gydF4y2Ba
는gydF4y2BaMdlgydF4y2Ba
,gydF4y2BaHyperparameterOptimizationResultsgydF4y2Ba
) = fitcecoc (gydF4y2Ba___gydF4y2Ba,gydF4y2Ba名称,值gydF4y2Ba
)gydF4y2BaOptimizeHyperparametersgydF4y2Ba
이름——값쌍의인수를지정하고선형또는커널이진학습기를사용할경우하이퍼파라미터최적화세부정보도반환합니다。다른gydF4y2Ba学习者gydF4y2Ba
의경우,gydF4y2BaMdlgydF4y2Ba
의gydF4y2BaHyperparameterOptimizationResultsgydF4y2Ba
속성에결과가들어있습니다。gydF4y2Ba
서포트벡터머신(SVM)이진학습기를사용하여다중클래스오류수정출력코드(ECOC)모델을훈련시킵니다。gydF4y2Ba
피셔(费舍尔)의붓꽃데이터세트를불러옵니다。예측변수데이터gydF4y2BaXgydF4y2Ba
와응답변수데이터gydF4y2BaYgydF4y2Ba
를 지정합니다.gydF4y2Ba
负载gydF4y2BafisheririsgydF4y2BaX =量;Y =物种;gydF4y2Ba
디폴트옵션을사용하여다중클래스ECOC모델을훈련시킵니다。gydF4y2Ba
Mdl = fitcecoc (X, Y)gydF4y2Ba
Mdl = ClassificationECOC ResponseName: 'Y' CategoricalPredictors: [] ClassNames: {'setosa' 'versicolor' 'virginica'} ScoreTransform: 'none' BinaryLearners: {3x1 cell} CodingName: 'onevsone'属性,方法gydF4y2Ba
MdlgydF4y2Ba
은gydF4y2BaClassificationECOCgydF4y2Ba
모델입니다。기본적으로,gydF4y2BafitcecocgydF4y2Ba
는SVM이진학습기와일대일코딩설계를사용합니다。점표기법을사용하여gydF4y2BaMdlgydF4y2Ba
속성에액세스할수있습니다。gydF4y2Ba
클래스이름과코딩설계행렬을표시합니다。gydF4y2Ba
Mdl。Cl一个年代年代的名字年代gydF4y2Ba
ans =gydF4y2Ba3 x1细胞gydF4y2Ba{'setosa'} {'versicolor'} {'virginica'}gydF4y2Ba
CodingMat = Mdl。CodingMatrixgydF4y2Ba
CodingMat =gydF4y2Ba3×3gydF4y2Ba1 1 0 1 0 1 1gydF4y2Ba
3.개 클래스에 대해 일대일 코딩 설계를 하면 3.개의 이진 학습기가 생성됩니다.gydF4y2BaCodingMatgydF4y2Ba
의열은학습기에대응되고행은클래스에대응됩니다。클래스순서는gydF4y2BaMdl。Cl一个年代年代的名字年代gydF4y2Ba
의순서와일치합니다。예를들어,gydF4y2BaCodingMat (: 1)gydF4y2Ba
은gydF4y2Ba[1; –1; 0]gydF4y2Ba
이며,gydF4y2Ba“塞托萨”gydF4y2Ba
와gydF4y2Ba“多色的”gydF4y2Ba
로분류된모든관측값을사용하여첫번째SVM이진학습기를훈련시켰음을나타냅니다。gydF4y2Ba“塞托萨”gydF4y2Ba
는gydF4y2Ba1gydF4y2Ba
에 대응되므로 양성 클래스이고,gydF4y2Ba“多色的”gydF4y2Ba
는gydF4y2Ba1gydF4y2Ba
에대응되므로음성클래스입니다。gydF4y2Ba
셀 인덱싱과 점 표기법을 사용하여 이진 학습기 각각에 액세스할 수 있습니다.gydF4y2Ba
Mdl。BinaryLearners {1}gydF4y2Ba第一个二元学习者gydF4y2Ba
ans = CompactClassificationSVM ResponseName: 'Y' CategoricalPredictors: [] ClassNames: [-1 1] ScoreTransform: 'none' Beta: [4x1 double] Bias: 1.4492 KernelParameters: [1x1 struct] Properties, Methods . txt: [gydF4y2Ba
재대입분류오차를계산합니다。gydF4y2Ba
错误= resubLoss (Mdl)gydF4y2Ba
错误= 0.0067gydF4y2Ba
훈련데이터에대한분류오차가작지만,분류기가과적합된모델일수있습니다。그대신gydF4y2BacrossvalgydF4y2Ba
을사용하여분류기를교차검증하여교차검증분류오차를계산할수있습니다。gydF4y2Ba
다중 이진 선형 분류 모델로 구성된 经济合作모델을 훈련시킵니다.gydF4y2Ba
NLP(자연어처리)데이터세트를불러옵니다。gydF4y2Ba
负载gydF4y2BanlpdatagydF4y2Ba
XgydF4y2Ba
는예측변수데이터로구성된희소행렬이고,gydF4y2BaYgydF4y2Ba
는클래스레이블로구성된直言형벡터입니다。이데이터에는세개이상의클래스가있습니다。gydF4y2Ba
디폴트선형분류모델템플릿을생성합니다。gydF4y2Ba
t = templateLinear ();gydF4y2Ba
디폴트값을조정하려면gydF4y2BatemplateLineargydF4y2Ba
페이지의gydF4y2Ba名称-值对的观点gydF4y2Ba항목을참조하십시오。gydF4y2Ba
문서웹페이지에나오는단어의도수분포가주어진경우,어떤제품에대한페이지인지를식별할수있는다중이진선형분류모델로구성된ECOC모델을훈련시킵니다。훈련시간을단축하도록예측변수데이터를전치하고관측값이열에대응됨을지정하십시오。gydF4y2Ba
X = X ';rng (1);gydF4y2Ba%为了再现性gydF4y2BaMdl = fitcecoc (X, Y,gydF4y2Ba“学习者”gydF4y2BatgydF4y2Ba“ObservationsIn”gydF4y2Ba,gydF4y2Ba“专栏”gydF4y2Ba)gydF4y2Ba
Mdl = CompactClassificationECOC ResponseName: 'Y' ClassNames: [1x13 categorical] ScoreTransform: 'none' BinaryLearners: {78x1 cell} CodingMatrix: [13x78 double]属性,方法gydF4y2Ba
또는,gydF4y2Ba“学习者”,“线性”gydF4y2Ba
를사용하여디폴트선형분류모델로구성된ECOC모델을훈련시킬수있습니다。gydF4y2Ba
메모리를절약하기위해gydF4y2BafitcecocgydF4y2Ba
는선형분류학습기로구성된훈련된ECOC모델을gydF4y2Ba紧凑分类gydF4y2Ba
모델객체로반환합니다。gydF4y2Ba
SVM이진학습기를사용하여ECOC분류기를교차검증하고일반화된분류오차를추정합니다。gydF4y2Ba
피셔(费舍尔)의붓꽃데이터세트를불러옵니다。예측변수데이터gydF4y2BaXgydF4y2Ba
와응답변수데이터gydF4y2BaYgydF4y2Ba
를 지정합니다.gydF4y2Ba
负载gydF4y2BafisheririsgydF4y2BaX =量;Y =物种;rng (1);gydF4y2Ba%为了再现性gydF4y2Ba
SVM템플릿을생성하고예측변수를표준화합니다。gydF4y2Ba
t=模板SVM(gydF4y2Ba“标准化”gydF4y2Ba,真正的)gydF4y2Ba
t =适合分类支持向量机的模板。α:[0 x1双]BoxConstraint: [] CacheSize: [] CachingMethod:“ClipAlphas: [] DeltaGradientTolerance:[]ε:[]GapTolerance: [] KKTTolerance: [] IterationLimit: [] KernelFunction:“KernelScale: [] KernelOffset: [] KernelPolynomialOrder: [] NumPrint:[]ν:[]OutlierFraction: [] RemoveDuplicates: [] ShrinkagePeriod:[]解算器:"标准化数据:1 SaveSupportVector万博1manbetxs: [] VerbosityLevel:[]版本:2方法:'SVM'类型:'分类'gydF4y2Ba
tgydF4y2Ba
는SVM템플릿입니다。템플릿객체의속성대부분은비어있습니다。ECOC분류기를훈련시킬때소프트웨어는적합한속성을해당디폴트값으로설정합니다。gydF4y2Ba
ECOC분류기를훈련시키고클래스순서를지정합니다。gydF4y2Ba
Mdl = fitcecoc (X, Y,gydF4y2Ba“学习者”gydF4y2BatgydF4y2Ba...gydF4y2Ba“类名”gydF4y2Ba,{gydF4y2Ba“塞托萨”gydF4y2Ba,gydF4y2Ba“多色的”gydF4y2Ba,gydF4y2Ba“维吉尼亚”gydF4y2Ba});gydF4y2Ba
MdlgydF4y2Ba
은gydF4y2BaClassificationECOCgydF4y2Ba
분류기입니다。점표기법을사용하여해당속성에액세스할수있습니다。gydF4y2Ba
10겹교차검증을사용하여gydF4y2BaMdlgydF4y2Ba
을교차검증합니다。gydF4y2Ba
CVMdl = crossval (Mdl);gydF4y2Ba
CVMdlgydF4y2Ba
은교차검증된ECOC분류기gydF4y2BaClassificationPartitionedECOCgydF4y2Ba
입니다。gydF4y2Ba
일반화된분류오차를추정합니다。gydF4y2Ba
genError = kfoldLoss (CVMdl)gydF4y2Ba
genError = 0.0400gydF4y2Ba
일반화된분류오차는4%이며,이는ECOC분류기가일반화를상당히잘함을나타냅니다。gydF4y2Ba
支持向量机이진 학습기를 사용하여 经济合作분류기를 훈련시킵니다. 먼저 훈련-표본 레이블과 클래스 사후 확률을 예측합니다. 그런 다음 그리드의 각 점에서 클래스 사후 확률 최댓값을 예측합니다. 결과를 시각화합니다.gydF4y2Ba
피셔(费舍尔)의붓꽃데이터세트를불러옵니다。꽃잎치수를예측변수로지정하고종이름을응답변수로지정합니다。gydF4y2Ba
负载gydF4y2BafisheririsgydF4y2BaX =量(:,3:4);Y =物种;rng (1);gydF4y2Ba%为了再现性gydF4y2Ba
SVM템플릿을생성합니다。예측변수를표준화하고가우스커널을지정합니다。gydF4y2Ba
t=模板SVM(gydF4y2Ba“标准化”gydF4y2Ba,真的,gydF4y2Ba“KernelFunction”gydF4y2Ba,gydF4y2Ba“高斯”gydF4y2Ba);gydF4y2Ba
tgydF4y2Ba
는SVM템플릿입니다。해당속성대부분이비어있습니다。소프트웨어는ECOC분류기를훈련시킬때적합한속성을해당디폴트값으로설정합니다。gydF4y2Ba
SVM템플릿을사용하여ECOC분류기를훈련시킵니다。gydF4y2Ba“FitPosterior”gydF4y2Ba
이름——값쌍의인수를사용하여분류점수를클래스사후확률(gydF4y2Ba预测gydF4y2Ba
또는gydF4y2BaresubPredictgydF4y2Ba
에서반환됨)로변환합니다。gydF4y2Ba“类名”gydF4y2Ba
이름——값쌍의인수를사용하여클래스순서를지정합니다。gydF4y2Ba“详细”gydF4y2Ba
이름——값쌍의인수를사용하여훈련중에진단메시지를표시합니다。gydF4y2Ba
Mdl = fitcecoc (X, Y,gydF4y2Ba“学习者”gydF4y2BatgydF4y2Ba“FitPosterior”gydF4y2Ba,真的,gydF4y2Ba...gydF4y2Ba“类名”gydF4y2Ba,{gydF4y2Ba“塞托萨”gydF4y2Ba,gydF4y2Ba“多色的”gydF4y2Ba,gydF4y2Ba“维吉尼亚”gydF4y2Ba},gydF4y2Ba...gydF4y2Ba“详细”gydF4y2Ba,2);gydF4y2Ba
训练二元学习者1(SVM),共3个,分别有50个负面观察和50个正面观察。负面类指数:2个正面类指数:1个拟合学习者1(SVM)的后验概率。训练二元学习者2(SVM)三分之三有50个负面和50个正面观察结果。负面类指数:3个正面类指数:1拟合学习者2的后验概率(SVM)。在三分之三有50个负面和50个正面观察结果的二元学习者中培训三分之三(SVM)。负面类指数:3个正面类指数:2拟合学习者3的后验概率(SVM)。gydF4y2Ba
MdlgydF4y2Ba
은gydF4y2BaClassificationECOCgydF4y2Ba
모델입니다。동일한SVM템플릿이각각의이진학습기에적용되지만,템플릿으로구성된셀형벡터를전달하여이진학습기별로옵션을조정할수있습니다。gydF4y2Ba
훈련——표본레이블과클래스사후확률을예측합니다。gydF4y2Ba“详细”gydF4y2Ba
이름——값쌍의인수를사용하여레이블과클래스사후확률을계산하는중에진단메시지를표시합니다。gydF4y2Ba
[标签,~,~,后]= resubPredict (Mdl,gydF4y2Ba“详细”gydF4y2Ba1);gydF4y2Ba
计算所有学习者的预测。计算所有观察的损失。计算后验概率。。。gydF4y2Ba
Mdl.BinaryLossgydF4y2Ba
ans =“二次”gydF4y2Ba
소프트웨어는가장작은평균이진손실을생성하는클래스에관측값을할당합니다。모든이진학습기가사후확률을계산하기때문에이진손실함수는gydF4y2Ba二次的gydF4y2Ba
입니다。gydF4y2Ba
임의의결과로구성된집합을표시합니다。gydF4y2Ba
idx = randsample(大小(X, 1), 10日1);Mdl。Cl一个年代年代的名字年代gydF4y2Ba
ans =gydF4y2Ba3 x1细胞gydF4y2Ba{'setosa'} {'versicolor'} {'virginica'}gydF4y2Ba
表(Y (idx)、标签(idx)、后(idx:)gydF4y2Ba...gydF4y2Ba“VariableNames”gydF4y2Ba,{gydF4y2Ba“TrueLabel”gydF4y2Ba,gydF4y2Ba“预标签”gydF4y2Ba,gydF4y2Ba“后部”gydF4y2Ba})gydF4y2Ba
ans =gydF4y2Ba10×3表gydF4y2BaTrueLabel PredLabel后 ______________ ______________ ______________________________________ {' virginica’}{‘virginica} 0.0039322 0.003987 0.99208{‘virginica}{‘virginica} 0.017067 0.018263 0.96467{‘virginica}{‘virginica} 0.014948 0.015856 0.9692{“癣”}{“癣”}2.2197 e-14 0.87318 - 0.12682{‘setosa} {' setosa '}0.999 0.029985 {'versicolor'} {'versicolor'}} 0.0085642 0.98259 0.0088487 {'setosa'} {'setosa'} 0.999 0.00024992 0.0088487 {'setosa'} {'setosa'} 0.999 0.00024913 0.00074717gydF4y2Ba
后gydF4y2Ba
의 열은gydF4y2BaMdl。Cl一个年代年代的名字年代gydF4y2Ba
의클래스순서에대응됩니다。gydF4y2Ba
관측된예측변수공간에서값의그리드를정의합니다。그리드의각인스턴스에대해사후확률을예측합니다。gydF4y2Ba
xMax=max(X);xMin=min(X);x1Pts=linspace(xMin(1),xMax(1));x2Pts=linspace(xMin(2),xMax(2));[x1Grid,x2Grid]=meshgrid(x1Pts,x2Pts);[~,~,~,PosteriorRegion]=predict(Mdl,[x1Grid(:),x2Grid(:)];gydF4y2Ba
그리드의 각 좌표에 대해 모든 클래스 중에서 최대 클래스 사후 확률을 플로팅합니다.gydF4y2Ba
contourf (x1Grid x2Grid,gydF4y2Ba...gydF4y2Ba重塑(max (PosteriorRegion[], 2),大小(x1Grid, 1),大小(x1Grid, 2)));h = colorbar;h.YLabel.String =gydF4y2Ba最大后验的gydF4y2Ba;h.YLabel.FontSize = 15;持有gydF4y2Ba在gydF4y2Bagh = gscatter (X (: 1), X (:, 2), Y,gydF4y2Ba“krk”gydF4y2Ba,gydF4y2Ba‘* xd‘gydF4y2Ba8);gh(2)。l我neWidth = 2; gh(3).LineWidth = 2; title(“虹膜瓣测量和最大后角”gydF4y2Ba)包含(gydF4y2Ba“花瓣长度(厘米)”gydF4y2Ba) ylabel (gydF4y2Ba“花瓣宽度(cm)”gydF4y2Ba)轴gydF4y2Ba紧gydF4y2Ba传奇(gh,gydF4y2Ba“位置”gydF4y2Ba,gydF4y2Ba“西北”gydF4y2Ba)持有gydF4y2Ba从gydF4y2Ba
다음 제품이 필요합니다.gydF4y2Ba
대리 분할을 활용하는 결정 트리로 구성된gydF4y2BaGentleBoostgydF4y2Ba
앙상블을사용하여일대전부ECOC분류기를훈련시킵니다。훈련속도를높이기위해숫자형예측변수를비닝하고병렬연산을사용합니다。비닝은gydF4y2BafitcecocgydF4y2Ba
가트리학습기를사용하는경우에만유효합니다。훈련후에10겹교차검증을사용하여분류오차를추정합니다。병렬연산을수행하려면并行计算工具箱™가필요합니다。gydF4y2Ba
표본데이터를불러오기gydF4y2Ba
心律失常gydF4y2Ba
데이터세트를불러온후살펴봅니다。gydF4y2Ba
负载gydF4y2Ba心律失常gydF4y2Ba(氮、磷)大小(X) =gydF4y2Ba
n = 452gydF4y2Ba
p = 279gydF4y2Ba
isLabels=unique(Y);nLabels=numel(isLabels)gydF4y2Ba
nLabels = 13gydF4y2Ba
汇总(分类(Y))gydF4y2Ba
价值计数百分比124554.20%2449.73%3153.32%4153.32%5132.88%6255.53%730.66%820.44%991.99%1050111.06%1440.88%1551.11%16224.87%gydF4y2Ba
데이터세트에는gydF4y2Ba279gydF4y2Ba
개예측변수가있으며,표본크기는gydF4y2Ba452gydF4y2Ba
개로상대적으로작습니다。16개의고유한레이블중13개만응답변수(gydF4y2BaYgydF4y2Ba
)에나타납니다。각레이블은다양한부정맥의정도를나타내며,관측값가중54.20%클래스gydF4y2Ba1gydF4y2Ba
에있습니다。gydF4y2Ba
일대전부ECOC분류기훈련시키기gydF4y2Ba
앙상블템플릿을생성합니다。적어도세개의인수,방법,학습기개수및학습기유형을지정해야합니다。이예에서는방법으로gydF4y2Ba“绅士之声”gydF4y2Ba
,학습기개수로gydF4y2BaOne hundred.gydF4y2Ba
을지정하고,누락된관측값이있으므로대리분할을사용하는결정트리템플릿을지정하십시오。gydF4y2Ba
tTree = templateTree (gydF4y2Ba“代孕”gydF4y2Ba,gydF4y2Ba“上”gydF4y2Ba);tEnsemble = templateEnsemble (gydF4y2Ba“绅士之声”gydF4y2Ba,100,t树);gydF4y2Ba
tEnsemblegydF4y2Ba
은템플릿객체입니다。해당속성의대부분은비어있지만,소프트웨어가훈련중에해당디폴트값으로속성을채웁니다。gydF4y2Ba
결정트리로구성된앙상블을이진학습기로사용하여일대전부ECOC분류기를훈련시킵니다。훈련속도를높이기위해비닝과병렬연산을사용합니다。gydF4y2Ba
비닝(gydF4y2Ba“NumBins”,50岁gydF4y2Ba
)——훈련데이터세트가큰경우gydF4y2Ba“NumBins”gydF4y2Ba
이름——값쌍의인수를사용하여훈련속도를높일수있습니다(이경우정확도가떨어질수있음)。이인수는gydF4y2BafitcecocgydF4y2Ba
가트리학습기를사용하는경우에만유효합니다。gydF4y2Ba“NumBins”gydF4y2Ba
값을지정하면모든숫자형예측변수를지정된개수의등확률本으로비닝한다음원래데이터가아닌本인덱스에서트리를성장시킵니다。먼저gydF4y2Ba“NumBins”,50岁gydF4y2Ba
을 사용해 본 후에 정확도와 훈련 속도에 따라gydF4y2Ba“NumBins”gydF4y2Ba
를변경해볼수있습니다。gydF4y2Ba
병렬 연산(gydF4y2Ba“选项”,statset('UseParallel',true)gydF4y2Ba
)——并行计算工具箱라이선스가있는경우각각의이진학습기를풀에있는각워커로전송하는병렬연산을사용하여계산속도를높일수있습니다。워커개수는시스템구성에따라달라집니다。이진학습기에결정트리를사용할때는gydF4y2BafitcecocgydF4y2Ba
는듀얼코어이상의시스템에서英特尔®TBB(线程构建块)를사용하여훈련을병렬화합니다。따라서단일컴퓨터에서gydF4y2Ba“UseParallel”gydF4y2Ba
옵션을지정하는것은도움이되지않습니다。이옵션은클러스터에서사용하십시오。gydF4y2Ba
또한,사전확률이1 /gydF4y2BaKgydF4y2Ba임을지정합니다。여기서gydF4y2BaKgydF4y2Ba= 13은고유한클래스개수입니다。gydF4y2Ba
选择= statset (gydF4y2Ba“UseParallel”gydF4y2Ba,真正的);Mdl = fitcecoc (X, Y,gydF4y2Ba“编码”gydF4y2Ba,gydF4y2Ba“onevsall”gydF4y2Ba,gydF4y2Ba“学习者”gydF4y2BatEnsemble,gydF4y2Ba...gydF4y2Ba“之前”gydF4y2Ba,gydF4y2Ba“统一”gydF4y2Ba,gydF4y2Ba“NumBins”gydF4y2Ba, 50岁,gydF4y2Ba“选项”gydF4y2Ba、选择);gydF4y2Ba
使用“local”配置文件启动并行池(parpool)…连接到并行池(工作人员数量:6)。gydF4y2Ba
MdlgydF4y2Ba
은gydF4y2BaClassificationECOCgydF4y2Ba
모델입니다。gydF4y2Ba
교차검증gydF4y2Ba
10겹교차검증을사용하여ECOC분류기를교차검증합니다。gydF4y2Ba
CVMdl = crossval (Mdl,gydF4y2Ba“选项”gydF4y2Ba、选择);gydF4y2Ba
警告:一个或多个折叠不包含所有组中的点。gydF4y2Ba
CVMdlgydF4y2Ba
은gydF4y2BaClassificationPartitionedECOCgydF4y2Ba
모델입니다。경고는소프트웨어가최소1겹을훈련시키는데일부클래스가사용되지않았음을나타냅니다。따라서,이러한겹은누락된클래스에대한레이블을예측할수없습니다。사용자는셀인덱싱과점표기법을사용하여겹의결과를검사할수있습니다。예를들어,gydF4y2BaCVMdl。训练有素的{1}gydF4y2Ba
을입력하여첫번째겹의결과에액세스할수있습니다。gydF4y2Ba
교차검증된ECOC분류기를사용하여검증——겹레이블을예측합니다。gydF4y2BaconfusionchartgydF4y2Ba
를사용하여정오분류표를계산할수있습니다。내부위치속성을변경하여차트를이동하고크기조정하여행요약에백분율이나타나도록합니다。gydF4y2Ba
oofLabel = kfoldPredict (CVMdl,gydF4y2Ba“选项”gydF4y2Ba,选项);ConfMat=confusionchart(Y,oofLabel,gydF4y2Ba“RowSummary”gydF4y2Ba,gydF4y2Ba“总标准化”gydF4y2Ba);ConfMat。InnerPosition = [0.10 0.12 0.85 0.85];gydF4y2Ba
비닝된데이터재현하기gydF4y2Ba
훈련된 모델의gydF4y2BaBinEdgesgydF4y2Ba
속성과gydF4y2Ba离散化gydF4y2Ba
함수를 사용하여, 비닝된 예측 변수 데이터를 재현합니다.gydF4y2Ba
X = Mdl.X;gydF4y2Ba%的预测数据gydF4y2BaXbinned = 0(大小(X));边缘= Mdl.BinEdges;gydF4y2Ba找到被分类的预测器的指数。gydF4y2BaidxNumeric =找到(~ cellfun (@isempty边缘));gydF4y2Ba如果gydF4y2Baiscolumn(idxNumeric) idxNumeric = idxNumeric';gydF4y2Ba结束gydF4y2Ba为gydF4y2Baj=idxNumeric x=x(:,j);gydF4y2Ba%如果x是一个表,则将x转换为数组。gydF4y2Ba如果gydF4y2BaIstable (x) x = table2array(x);gydF4y2Ba结束gydF4y2Ba%使用离散函数将x分组到bins中。gydF4y2Baxbinned =离散化(x,[无穷;边缘{};正]);Xbinned (:, j) = Xbinned;gydF4y2Ba结束gydF4y2Ba
XbinnedgydF4y2Ba
는숫자형예측변수에대해1부터本개수사이의本인덱스를포함합니다。범주형예측변수의경우gydF4y2BaXbinnedgydF4y2Ba
값은gydF4y2Ba0gydF4y2Ba
입니다。gydF4y2BaXgydF4y2Ba
가gydF4y2Ba南gydF4y2Ba
을포함하는경우이에대응되는gydF4y2BaXbinnedgydF4y2Ba
값은gydF4y2Ba南gydF4y2Ba
이됩니다。gydF4y2Ba
fitcecocgydF4y2Ba
를 사용하여 하이퍼파라미터를 자동으로 최적화합니다.gydF4y2Ba
fisheririsgydF4y2Ba
데이터세트를불러옵니다。gydF4y2Ba
负载gydF4y2BafisheririsgydF4y2BaX =量;Y =物种;gydF4y2Ba
자동 하이퍼파라미터 최적화를 사용하여 5.겹 교차 검증 손실을 최소화하는 하이퍼파라미터를 구합니다. 재현이 가능하도록 난수 시드값을 설정하고gydF4y2Ba“expected-improvement-plus”gydF4y2Ba
획득함수를사용합니다。gydF4y2Ba
rnggydF4y2Ba默认的gydF4y2BaMdl = fitcecoc (X, Y,gydF4y2Ba“OptimizeHyperparameters”gydF4y2Ba,gydF4y2Ba“汽车”gydF4y2Ba,gydF4y2Ba...gydF4y2Ba“HyperparameterOptimizationOptions”gydF4y2Ba结构(gydF4y2Ba“AcquisitionFunctionName”gydF4y2Ba,gydF4y2Ba...gydF4y2Ba“expected-improvement-plus”gydF4y2Ba))gydF4y2Ba
|第二个月第礼礼礼礼礼礼第第第第二礼礼礼礼第第第第第第礼礼礼礼第第第第第第礼礼礼第第第第礼礼礼第第第第第礼第第礼第第第礼第第第第礼第第礼第第第礼第第礼第第第礼第第礼第第礼第第第礼第第第第礼第第第第第礼第第第第第礼第第第第第礼第第第第第礼第第第第第第礼第第第第第第礼第第第第第礼第第第第礼第第第第第礼礼第第第第礼第第礼第第第礼第第第礼第第第礼第第第礼第第第第第第礼第第第第第第礼第第第第第第第第礼第第第第第第礼第第第第第第第礼第第第第第第第礼第第第第第第礼第第第第第第第第第第第第第第第运行时|(已观察)|(预计)第1244号?124;124124;124|124124|124|;124|124|124124|124124|;124124|;124124|124124|;124124124124124|124124124|;;第四四四四四四四四月月月月第四四四四四四方方方方第第第第第第第四方方方第第第第第第第第第四方方第第第第第第第第第四方方第第第第第四方方第第第第第第第第第第四方方第第三方方第第第第第第第三方第第第第第第第第第第第第第第第四方方第第第7 7 7===================================================39 | 200.36 | 2 |最佳| 0.066667 | 3.6612 | 0.066667 | 0.068735 |一对一| 94.849 | 0.0032549 | 3 |接受| 0.08 | 0.4554 | 0.066667 1246837 |“0.24984 0.0 0.24周四周四周四周四周四周四周四周四周四周四0.24周四周四周四周四周四0.0 0 0.24984周四周四周四周四0.0.06666712400.0.0.06667 0.0 0 0.06667 0.0 0 0.01013778 0.0137780.01780 0 0.0178周日周日周日周日周日周日周日周日周日0 0 0 0 0.01周日周日周日周日周日周日0.0 0 0 0 0 0.06六六六六六六六六六六六六六七0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0.06六六六六六六六六六六六六六六六六0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0.04 | 0.039999 | onevsone | 0.48338 | 0.02941 | 7 |接受| 0.04 | 0.33704 | 0.04 | 0.039989 | onevsone | 305.45 |0.0.066667124470.18647周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四0 0 0 0.0.0.0.186周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四周四124; 0.026673 | onevsone | 736.18 | 0.071026 | 11 |接受| 0.04 | 0.35546 | 0.026667 | 0.035679 | onevsone | 35.928|0.01774 0.026667 1240.0 0.030060 0.0 0 0.0 0 0 0 0.3 0 0 0 0 0 0 0 0.3 0 0 0 0.0 0 0 0.0 0 0 0.0 0 0 0 0 0.0 0 0 0 0.0 0 0 0.017912479124九九九九九九九九九九九九九九九九九九九九九九九四四四四四四四四四四0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0.0.3 0 0 0 0 0 0 0 0 0 0 0 0 0 0.3 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0.0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0.0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0| 0.026089 | onevsone | 0.0011124 | 0.079161 | 15 |接受| 0.026667 | 0.24303 | 0.026667 | 0.026184 |0.02 1240.2070.0 0 0.20712400.0 0 0.0 0 0.21240.0 0 0 0.0 0 0 0.21240.0 0 0.0 0 0 0.02124412400 0.0 0 0.01244 0.0 0 0.0010.0 0 0.0 0 0 0 0.0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 21244124412400 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 2124卡卡卡卡卡卡四四四四四四四四四四四四四四四四四四四四四四四四0.02 | 0.024292 | onevsone | 0.0011889 | 0.02915 | 19 |接受| 0.02 | 0.31439 | 0.02 | 0.022327 | onevsone |第0.0.01333 0.0.013 3 3 3 3 0.0.013 3 3 3 0.3 3 3 3 3 0.0 0.013 3 3 3 3 0.0.013 3 3 3 3 3 0.0.013 3 3 3 3 3 0.0.0.01133 3 3 0.0 0.0 0.0 0 0 0 0 0.0 0 0 0 0 0 0 0 0 0.0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0===============国际热核实验堆(Iter)评估(Eval)目标(Objective)最佳状态(BestSoFar)编码(BoxConstraint)核尺度(KernelScale)结果(result)运行时|(已观察)|(预计)第1244号?|;124124;124|124|124|;124124|124|124|124124|;124|124|124124|124124|124124124|;124124124|;124124|;;第四四四四四四四四四四月月月月月第四四四四四四四四号????????\124124124124;124|||124|||||;|;|;|;||124|;;|;;;|第第四四四四四号号号号号号号|0.001007 | 22 |接受| 0.33333 | 0.33577 | 0.013333 | 0.018299 | onevsall | 0.0011091 | 1.2155 | 23 |接受| 0.33333 | 0.38918 | 0.013330.017851 0.01780 0.01785112400.0178511240.0178511240.0178511240.01780.0178511240.0178511240.0178511240.01780.0178511240.0178511240.0178511240.181244.181244.18| 24124四四四四四四四四)接受接受;接受;241244四四四四四四个接受;接受;接受;接受;接受;接受0.0.040.040.0.04六六六六六六六六六7 0.0.0.0.0.0.0.0.0.0466671240.0.0.414141671240.0.0.4141414141414141671240.0.0 0 0 0 0 0.41414141414141414167| 0.31981 | 0.01333 | 0.018226 | onevsone | 0.0010775 | 999.54 | 27 |接受| 0.04 | 0.26125 | 0.01333 |0.01857 0.0.018557 0.0185571240.0.0185571240.0 0.01855712400.0185571240.0.0.01855 0.0.0185571240.0 0.018571240.0.0185571240.0 0.01857 0.0 0 0 0.0 0 0 0 0.0 0 0 0 0 0 0 0 0 0 0 0 0.01855 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5五五五五五五五五五五五五五五五五五五五五五五0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 04 | 0.24213 | 0.01333 | 0.018597 | onevsone | 936.87 | 1.7813|gydF4y2Ba
__________________________________________________________ 优化完成。maxobjective达到30个。总函数计算:30总运行时间:62.709秒总目标函数计算时间:39.5724最佳观测可行点:编码框约束KernelScale ________ _____________ ___________ onevsone 0.0010854 0.048345观测目标函数值= 0.013333估计目标函数值= 0.018594函数计算时间= 0.32092最佳估计可行点(根据模型):编码框约束KernelScale ________ _____________ ___________ onevsone 0.0011336 0.042445估计的目标函数值= 0.018597估计的函数计算时间= 0.2867gydF4y2Ba
Mdl = ClassificationECOC ResponseName: 'Y' CategoricalPredictors: [] ClassNames: {'setosa' 'versicolor' 'virginica'} ScoreTransform: 'none' BinaryLearners: {3x1 cell} CodingName: 'onevsone' HyperparameterOptimizationResults: [1x1 BayesianOptimization] Properties, Methods . class . php . php . php . php . php . php . php . php . php . php . php . php . php . php . php . php . php . php . php . php . php . phpgydF4y2Ba
高형데이터에대해훈련된다중클래스ECOC모델두개를생성합니다。한모델에는선형이진학습기를사용하고다른모델에는커널이진학습기를사용합니다。두모델에대한재대입분류오차를비교합니다。gydF4y2Ba
일반적으로,선형이진학습기또는커널이진학습기에gydF4y2BafitcecocgydF4y2Ba
를사용하여高형데이터에대한다중클래스분류를수행할수있습니다。gydF4y2BafitcecocgydF4y2Ba
를 사용하여 高的형 배열에 대해 모델을 훈련시키는 경우 支持向量机이진 학습기를 직접 사용할 수는 없습니다. 그러나 支持向量机을 사용하는 선형 이진 분류 모델 또는 커널 이진 분류 모델을 사용할 수는 있습니다.gydF4y2Ba
高형배열에대한계산을수행할때MATLAB®은병렬풀(并行计算工具箱™를사용할경우디폴트값)또는로컬MATLAB세션을사용합니다。并行计算工具箱가있는상태에서로컬MATLAB세션을사용하여예제를실행하려는경우gydF4y2BamapreducegydF4y2Ba
함수를사용하여전역실행환경을변경하면됩니다。gydF4y2Ba
피셔의 붓꽃 데이터 세트가 있는 폴더를 참조하는 데이터저장소를 생성합니다.gydF4y2Ba“NA”gydF4y2Ba
값을누락된데이터로지정하여gydF4y2Ba数据存储gydF4y2Ba
가이값을gydF4y2Ba南gydF4y2Ba
값으로대체하도록합니다。예측변수데이터및응답변수데이터를高형으로생성합니다。gydF4y2Ba
ds =数据存储(gydF4y2Ba“fisherris.csv”gydF4y2Ba,gydF4y2Ba“TreatAsMissing”gydF4y2Ba,gydF4y2Ba“NA”gydF4y2Ba);t =高(ds);gydF4y2Ba
使用“local”配置文件启动并行池(parpool)…连接到并行池(工作人员数量:6)。gydF4y2Ba
X = [t。年代epalLength t.SepalWidth t.PetalLength t.PetalWidth]; Y = t.Species;
예측변수데이터를표준화합니다。gydF4y2Ba
Z = zscore (X);gydF4y2Ba
高的형 데이터와 선형 이진 학습기를 사용하는 다중클래스 经济合作모델을 훈련시킵니다. 기본적으로, 高的형 배열을gydF4y2BafitcecocgydF4y2Ba
에전달하면소프트웨어는SVM을사용하는선형이진학습기를훈련시킵니다。응답변수데이터에고유한클래스가세개만있기때문에코딩체계를일대전부(高형데이터를사용하는경우디폴트값임)에서일대일(메모리내데이터를사용하는경우디폴트값임)로변경합니다。gydF4y2Ba
재현이 가능하도록gydF4y2BarnggydF4y2Ba
및gydF4y2BatallrnggydF4y2Ba
를사용하여난수생성기의시드값을설정합니다。결과는워커의개수및高형배열의실행환경에따라다를수있습니다。자세한내용은gydF4y2Ba控制代码运行的位置gydF4y2Ba항목을참조하십시오。gydF4y2Ba
rng(gydF4y2Ba“默认”gydF4y2Ba) tallrng (gydF4y2Ba“默认”gydF4y2Ba) mdlLinear = fitcecoc(Z,Y,gydF4y2Ba“编码”gydF4y2Ba,gydF4y2Ba“onevsone”gydF4y2Ba)gydF4y2Ba
训练二元学习者1(线性)3。训练二元学习者2(线性)3。训练二元学习者3(线性)的3。gydF4y2Ba
mdlLinear = CompactClassificationECOC ResponseName: 'Y' ClassNames: {'setosa' 'versicolor' 'virginica'} ScoreTransform: 'none' BinaryLearners: {3×1 cell} CodingMatrix: [3×3 double]属性,方法gydF4y2Ba
mdlLineargydF4y2Ba
는 세 개의 이진 학습기로 구성된gydF4y2Ba紧凑分类gydF4y2Ba
모델입니다。gydF4y2Ba
高형데이터와커널이진학습기를사용하는다중클래스ECOC모델을훈련시켜보겠습니다。먼저,gydF4y2Ba模板核gydF4y2Ba
객체를생성하여커널이진학습기의속성을지정합니다。특히,확장크기수를gydF4y2Ba
으로늘립니다。gydF4y2Ba
tKernel = templateKernel (gydF4y2Ba“NumExpansionDimensions”gydF4y2Ba2 ^ 16)gydF4y2Ba
tKernel =适合Kernel分类的模板。betaterance: [] BlockSize: [] BoxConstraint: [] Epsilon: [] NumExpansionDimensions: 65536 GradientTolerance: [] HessianHistorySize: [] IterationLimit: [] KernelScale: [] Lambda: [] Learner: 'svm' LossFunction: [] Stream: [] VerbosityLevel: [] Version: 1 Method: 'Kernel' Type: 'classification'gydF4y2Ba
기본적으로,커널이진학습기는SVM을사용합니다。gydF4y2Ba
模板核gydF4y2Ba
객체를gydF4y2BafitcecocgydF4y2Ba
로전달하고코딩체계를일대일로변경합니다。gydF4y2Ba
mdlKernel = fitcecoc (Z, Y,gydF4y2Ba“学习者”gydF4y2BatKernel,gydF4y2Ba“编码”gydF4y2Ba,gydF4y2Ba“onevsone”gydF4y2Ba)gydF4y2Ba
训练二进制学习者1(内核)3。训练二进制学习者2(内核)的3。训练二进制学习者3(内核)的3。gydF4y2Ba
mdlKernel = CompactClassificationECOC ResponseName: 'Y' ClassNames: {'setosa' 'versicolor' 'virginica'} ScoreTransform: 'none' BinaryLearners: {3×1 cell} CodingMatrix: [3×3 double]属性,方法gydF4y2Ba
mdlKernelgydF4y2Ba
도세개의이진학습기로구성된gydF4y2Ba紧凑分类gydF4y2Ba
모델입니다。gydF4y2Ba
두모델에대한재대입분류오차를비교합니다。gydF4y2Ba
errorLinear =收集(损失(mdlLinear, Z, Y))gydF4y2Ba
使用Parallel Pool 'local'计算tall表达式gydF4y2Ba
errorLinear = 0.0333gydF4y2Ba
errorKernel =收集(损失(mdlKernel, Z, Y))gydF4y2Ba
使用Parallel Pool 'local'计算tall表达式gydF4y2Ba
errorKernel = 0.0067gydF4y2Ba
mdlKernelgydF4y2Ba
이오분류하는훈련데이터의비율이gydF4y2BamdlLineargydF4y2Ba
보다더작습니다。gydF4y2Ba
资源描述gydF4y2Ba
- - - - - -gydF4y2Ba표본 데이터gydF4y2Ba표본데이터로,테이블로지정됩니다。gydF4y2Ba资源描述gydF4y2Ba
의각행은하나의관측값에대응되고,각열은하나의예측변수에대응됩니다。선택적으로,gydF4y2Ba资源描述gydF4y2Ba
은응답변수에대해하나의추가열을포함할수있습니다。문자형벡터로구성된셀형배열이외의셀형배열과다중열변수는허용되지않습니다。gydF4y2Ba
资源描述gydF4y2Ba
이응답변수를포함하며gydF4y2Ba资源描述gydF4y2Ba
의나머지모든변수를예측변수로사용하려는경우gydF4y2BaResponseVarNamegydF4y2Ba
을사용하여응답변수를지정하십시오。gydF4y2Ba
资源描述gydF4y2Ba
이응답변수를포함하며gydF4y2Ba资源描述gydF4y2Ba
의나머지변수중일부만예측변수로사용하려는경우gydF4y2Ba公式gydF4y2Ba
를사용하여공식을지정하십시오。gydF4y2Ba
资源描述gydF4y2Ba
이응답변수를포함하지않는경우,gydF4y2BaYgydF4y2Ba
를사용하여응답변수를지정하십시오。응답변수의길이와gydF4y2Ba资源描述gydF4y2Ba
의행개수는동일해야합니다。gydF4y2Ba
데이터형:gydF4y2Ba表格gydF4y2Ba
ResponseVarNamegydF4y2Ba
- - - - - -gydF4y2Ba응답변수이름gydF4y2Ba资源描述gydF4y2Ba
에포함된변수이름gydF4y2Ba응답변수이름으로,gydF4y2Ba资源描述gydF4y2Ba
의 변수 이름으로 지정됩니다.gydF4y2Ba
ResponseVarNamegydF4y2Ba
은문자형벡터나字符串형스칼라로지정해야합니다。예를들어,응답변수gydF4y2BaYgydF4y2Ba
가gydF4y2Ba资源描述。YgydF4y2Ba
로저장된경우이를gydF4y2Ba“Y”gydF4y2Ba
로지정하십시오。이렇게하지않으면모델을훈련시킬때gydF4y2BaYgydF4y2Ba
를포함한gydF4y2Ba资源描述gydF4y2Ba
의모든열이예측변수로처리됩니다。gydF4y2Ba
응답변수는直言형배열,문자형배열,字符串형배열,논리형벡터또는숫자형벡터,문자형벡터로구성된셀형배열이어야합니다。gydF4y2BaYgydF4y2Ba
가문자형배열인경우,응답변수의각요소는배열의각행에대응되어야합니다。gydF4y2Ba
一会gydF4y2Ba
이름——값인수를사용하여클래스의순서를지정하는것이좋습니다。gydF4y2Ba
데이터형:gydF4y2Ba字符gydF4y2Ba
|gydF4y2Ba字符串gydF4y2Ba
公式gydF4y2Ba
- - - - - -gydF4y2Ba응답변수와예측변수의부분집합에대한설명모델gydF4y2Ba응답변수,그리고예측변수의부분집합에대한설명모델로,gydF4y2Ba“Y ~ x1 + x2 + x3”gydF4y2Ba
형식의문자형벡터나字符串형스칼라로지정됩니다。이형식에서gydF4y2BaYgydF4y2Ba
는응답변수를나타내고,gydF4y2Bax1gydF4y2Ba
,gydF4y2Bax2gydF4y2Ba
,gydF4y2Bax3gydF4y2Ba
은예측변수를나타냅니다。gydF4y2Ba
资源描述gydF4y2Ba
의일부변수를모델훈련에사용할예측변수로지정하려면식을사용하십시오。사용자가식을지정하면gydF4y2Ba资源描述gydF4y2Ba
의변수중해당gydF4y2Ba公式gydF4y2Ba
에표시되지않은변수는사용되지않습니다。gydF4y2Ba
식에포함되는변수이름은gydF4y2Ba资源描述gydF4y2Ba
에 포함된 변수 이름(gydF4y2BaTbl.Properties.VariableNamesgydF4y2Ba
)이면서동시에유효한MATLABgydF4y2Ba®gydF4y2Ba식별자여야합니다。gydF4y2BaisvarnamegydF4y2Ba
함수를사용하여gydF4y2Ba资源描述gydF4y2Ba
에포함된변수이름을확인할수있습니다。변수이름이유효하지않으면gydF4y2Bamatlab.lang.makeValidNamegydF4y2Ba
함수를사용하여변수이름을변환할수있습니다。gydF4y2Ba
데이터형:gydF4y2Ba字符gydF4y2Ba
|gydF4y2Ba字符串gydF4y2Ba
YgydF4y2Ba
- - - - - -gydF4y2Ba클래스레이블gydF4y2BaECOC모델이훈련되는클래스레이블로,直言형배열,문자형배열,字符串형배열,논리형벡터또는숫자형벡터,문자형벡터로구성된셀형배열로지정됩니다。gydF4y2Ba
YgydF4y2Ba
가문자형배열인경우,각요소는배열의각행에대응되어야합니다。gydF4y2Ba
YgydF4y2Ba
의길이와gydF4y2Ba资源描述gydF4y2Ba
또는gydF4y2BaXgydF4y2Ba
의행개수는동일해야합니다。gydF4y2Ba
一会gydF4y2Ba
이름——값쌍의인수를사용하여클래스순서를지정하는것이좋습니다。gydF4y2Ba
데이터형:gydF4y2Ba分类gydF4y2Ba
|gydF4y2Ba字符gydF4y2Ba
|gydF4y2Ba字符串gydF4y2Ba
|gydF4y2Ba逻辑gydF4y2Ba
|gydF4y2Ba单gydF4y2Ba
|gydF4y2Ba双重的gydF4y2Ba
|gydF4y2Ba细胞gydF4y2Ba
XgydF4y2Ba
- - - - - -gydF4y2Ba예측변수데이터gydF4y2Ba예측변수데이터로,비희소행렬이나희소행렬로지정됩니다。gydF4y2Ba
YgydF4y2Ba
의길이와gydF4y2BaXgydF4y2Ba
의관측값개수는동일해야합니다。gydF4y2Ba
XgydF4y2Ba
에 나오는 순서로 예측 변수의 이름을 지정하려면gydF4y2BaPredictorNamesgydF4y2Ba
이름——값쌍의인수를사용하십시오。gydF4y2Ba
참고gydF4y2Ba
선형분류학습기의경우,관측값이열에대응되도록gydF4y2BaXgydF4y2Ba
의방향을지정하고gydF4y2Ba“ObservationsIn”、“列”gydF4y2Ba
를지정하면최적화-실행시간을상당히줄일수있습니다。gydF4y2Ba
다른모든학습기의경우관측값이행에대응되도록gydF4y2BaXgydF4y2Ba
의방향을지정하십시오。gydF4y2Ba
fitcecocgydF4y2Ba
함수는선형분류모델을훈련시키는데에만희소행렬을지원합니다。gydF4y2Ba
데이터형:gydF4y2Ba双重的gydF4y2Ba
|gydF4y2Ba单gydF4y2Ba
참고gydF4y2Ba
南gydF4y2Ba
,빈문자형벡터(gydF4y2Ba”gydF4y2Ba
),빈字符串형(gydF4y2Ba""gydF4y2Ba
),gydF4y2Ba< >失踪gydF4y2Ba
,gydF4y2Ba<定义>gydF4y2Ba
요소는누락데이터로처리됩니다。gydF4y2BaYgydF4y2Ba
의누락값에대응되는gydF4y2BaXgydF4y2Ba
의행은제거됩니다。그러나,gydF4y2BaXgydF4y2Ba
의누락값에대한처리는이진학습기마다다릅니다。자세한내용은gydF4y2BafitcdiscrgydF4y2Ba
,gydF4y2BafitckernelgydF4y2Ba
,gydF4y2BafitcknngydF4y2Ba
,gydF4y2BafitclineargydF4y2Ba
,gydF4y2BafitcnbgydF4y2Ba
,gydF4y2BafitcsvmgydF4y2Ba
,gydF4y2BafitctreegydF4y2Ba
,gydF4y2BafitcensemblegydF4y2Ba
등 사용하는 이진 학습기의 훈련 함수를 참조하십시오. 관측값을 제거하면 훈련의 효력 또는 교차 검증 표본 크기가 줄어듭니다.gydF4y2Ba
선택적으로gydF4y2Ba名称,值gydF4y2Ba
인수가쉼표로구분되어지정됩니다。여기서gydF4y2Ba的名字gydF4y2Ba
은인수이름이고gydF4y2Ba价值gydF4y2Ba
는대응값입니다。gydF4y2Ba的名字gydF4y2Ba
은따옴표안에표시해야합니다。gydF4y2BaName1, Value1,…,的家gydF4y2Ba
과같이여러개의이름——값쌍의인수를어떤순서로든지정할수있습니다。gydF4y2Ba
‘学习者’、‘树’、‘编码’、‘onevsone’、‘CrossVal’、‘on’gydF4y2Ba
은모든이진학습기에대한결정트리,일대일코딩설계를사용하고10겹교차검증을구현하도록지정합니다。gydF4y2Ba
참고gydF4y2Ba
교차검증이름——값쌍의인수는gydF4y2Ba“OptimizeHyperparameters”gydF4y2Ba
이름——값쌍의인수와함께사용할수없습니다。gydF4y2Ba“OptimizeHyperparameters”gydF4y2Ba
에대한교차검증을수정하려면gydF4y2Ba“HyperparameterOptimizationOptions”gydF4y2Ba
이름-값 쌍의 인수를 사용해야만 합니다.gydF4y2Ba
编码gydF4y2Ba
- - - - - -gydF4y2Ba코딩설계gydF4y2Ba“onevsone”gydF4y2Ba
(디폴트 값) |gydF4y2Ba“allpairs”gydF4y2Ba
|gydF4y2Ba“binarycomplete”gydF4y2Ba
|gydF4y2Ba“denserandom”gydF4y2Ba
|gydF4y2Ba“onevsall”gydF4y2Ba
|gydF4y2Ba“顺序”gydF4y2Ba
|gydF4y2Ba“sparserandom”gydF4y2Ba
|gydF4y2Ba“ternarycomplete”gydF4y2Ba
|gydF4y2Ba숫자형행렬gydF4y2Ba코딩설계이름으로,gydF4y2Ba“编码”gydF4y2Ba
과함께숫자형행렬또는다음표에나와있는값이쉼표로구분되어지정됩니다。gydF4y2Ba
값gydF4y2Ba | 이진학습기개수gydF4y2Ba | 설명gydF4y2Ba |
---|---|---|
“allpairs”gydF4y2Ba 와gydF4y2Ba“onevsone”gydF4y2Ba |
K (K - 1) / 2gydF4y2Ba | 각각의이진학습기에서한클래스는양성이고,다른클래스는음성이며,나머지클래스는무시됩니다。이설계는모든클래스쌍할당조합을사용합니다。gydF4y2Ba |
“binarycomplete”gydF4y2Ba |
이설계는클래스를모든이진조합으로분할하며어떠한클래스도무시하지않습니다。각각의이진학습기에서모든클래스는gydF4y2Ba-1gydF4y2Ba 과gydF4y2Ba1gydF4y2Ba 중하나로할당되며,각조합에는적어도하나의양성클래스와음성클래스가있습니다。gydF4y2Ba |
|
“denserandom”gydF4y2Ba |
임의(하지만 대략적으로 10原木gydF4y2Ba2gydF4y2BaK)gydF4y2Ba | 각각의 이진 학습기에서 소프트웨어는 클래스를 양성 클래스 또는 음성 클래스로 임의로 할당하되 적어도 하나의 양성 및 음성 클래스가 있도록 합니다. 자세한 내용은gydF4y2Ba확률코딩설계행렬gydF4y2Ba항목을참조하십시오。gydF4y2Ba |
“onevsall”gydF4y2Ba |
KgydF4y2Ba | 각각의이진학습기에서한클래스는양성이고나머지는음성입니다。이설계는모든양성클래스할당조합을사용합니다。gydF4y2Ba |
“顺序”gydF4y2Ba |
K - 1gydF4y2Ba | 첫번째이진학습기에서는첫번째클래스가음성이고나머지는양성입니다。두번째이진학습기에서는처음두개의클래스가음성이고나머지는양성이되는식입니다。gydF4y2Ba |
“sparserandom”gydF4y2Ba |
임의(하지만 대략적으로 15原木gydF4y2Ba2gydF4y2BaK)gydF4y2Ba | 각각의이진학습기에서소프트웨어는클래스를각각0.25의확률로양성또는음성으로임의로할당하고0.5의확률로무시합니다。자세한내용은gydF4y2Ba확률코딩설계행렬gydF4y2Ba항목을참조하십시오。gydF4y2Ba |
“ternarycomplete”gydF4y2Ba |
이설계는클래스를모든삼진(三元)조합으로분할합니다。모든클래스는gydF4y2Ba0gydF4y2Ba ,gydF4y2Ba-1gydF4y2Ba ,gydF4y2Ba1gydF4y2Ba 중 하나로 할당되며, 각 조합에는 적어도 하나의 양성 클래스와 적어도 하나의 음성 클래스가 있습니다.gydF4y2Ba |
사용자지정코딩행렬을사용하여코딩설계를지정할수도있습니다。사용자지정코딩행렬K L×은행렬입니다。각행은클래스에대응되며,각열은이진학습기에대응됩니다。클래스순서는(행)gydF4y2Ba一会gydF4y2Ba
의순서에대응됩니다。다음지침에따라행렬을구성하십시오。gydF4y2Ba
사용자지정코딩행렬의각요소는gydF4y2Ba-1gydF4y2Ba
,gydF4y2Ba0gydF4y2Ba
또는gydF4y2Ba1gydF4y2Ba
이어야하고요,소의값은이분클래스할당에대응해야합니다。다음표는gydF4y2Ba编码(i, j)gydF4y2Ba
가나타내는의미,즉학습기gydF4y2BajgydF4y2Ba
가gydF4y2Ba我gydF4y2Ba
클래스의관측값에할당하는클래스를설명하고있습니다。gydF4y2Ba
값gydF4y2Ba | 이분클래스할당gydF4y2Ba |
---|---|
1gydF4y2Ba |
학습기gydF4y2BajgydF4y2Ba 가gydF4y2Ba我gydF4y2Ba 클래스에있는관측값을음성클래스로할당합니다。gydF4y2Ba |
0gydF4y2Ba |
훈련전에학습기gydF4y2BajgydF4y2Ba 는 데이터 세트에서gydF4y2Ba我gydF4y2Ba 클래스에있는관측값을제거합니다。gydF4y2Ba |
1gydF4y2Ba |
학습기gydF4y2BajgydF4y2Ba 가gydF4y2Ba我gydF4y2Ba 클래스에있는관측값을양성클래스로할당합니다。gydF4y2Ba |
모든열은gydF4y2Ba-1gydF4y2Ba
또는gydF4y2Ba1gydF4y2Ba
을하나이상포함해야합니다。gydF4y2Ba
我gydF4y2Ba
≠gydF4y2BajgydF4y2Ba
를만족하는모든열인덱스gydF4y2Ba我gydF4y2Ba
,gydF4y2BajgydF4y2Ba
에대해gydF4y2Ba编码(:,我)gydF4y2Ba
는gydF4y2Ba编码(:,j)gydF4y2Ba
와같을수없으며gydF4y2Ba编码(:,我)gydF4y2Ba
는gydF4y2Ba编码(:,j)gydF4y2Ba
와같을수없습니다。gydF4y2Ba
사용자 지정 코딩 행렬의 모든 행은 달라야 합니다.gydF4y2Ba
사용자 지정 코딩 설계 행렬의 형식에 대한 자세한 내용은gydF4y2Ba사용자지정코딩설계행렬gydF4y2Ba항목을참조하십시오。gydF4y2Ba
예:gydF4y2Ba“编码”、“ternarycomplete”gydF4y2Ba
데이터형:gydF4y2Ba字符gydF4y2Ba
|gydF4y2Ba字符串gydF4y2Ba
|gydF4y2Ba双重的gydF4y2Ba
|gydF4y2Ba单gydF4y2Ba
|gydF4y2Baint16gydF4y2Ba
|gydF4y2Baint32gydF4y2Ba
|gydF4y2Baint64gydF4y2Ba
|gydF4y2Baint8gydF4y2Ba
FitPosteriorgydF4y2Ba
- - - - - -gydF4y2Ba점수를사후확률로변환할지여부를나타내는플래그gydF4y2Ba假gydF4y2Ba
또는gydF4y2Ba0gydF4y2Ba
(디폴트 값) |gydF4y2Ba真正的gydF4y2Ba
또는gydF4y2Ba1gydF4y2Ba
점수를사후확률로변환할지여부를나타내는플래그로,gydF4y2Ba“FitPosterior”gydF4y2Ba
와함께gydF4y2Ba真正的gydF4y2Ba
(gydF4y2Ba1gydF4y2Ba
)또는gydF4y2Ba假gydF4y2Ba
(gydF4y2Ba0gydF4y2Ba
)가쉼표로구분되어지정됩니다。gydF4y2Ba
FitPosteriorgydF4y2Ba
가gydF4y2Ba真正的gydF4y2Ba
이면 소프트웨어가 이진-학습기 분류 점수를 사후 확률로 변환합니다.gydF4y2BakfoldPredictgydF4y2Ba
,gydF4y2Ba预测gydF4y2Ba
또는gydF4y2BaresubPredictgydF4y2Ba
를사용하여사후확률을얻을수있습니다。gydF4y2Ba
fitcecocgydF4y2Ba
는다음과같은경우사후확률피팅을지원하지않습니다。gydF4y2Ba
앙상블방법이gydF4y2BaAdaBoostM2gydF4y2Ba
,gydF4y2BaLPBoostgydF4y2Ba
,gydF4y2BaRUSBoostgydF4y2Ba
,gydF4y2BaRobustBoostgydF4y2Ba
또는gydF4y2BaTotalBoostgydF4y2Ba
입니다。gydF4y2Ba
이진학습기(gydF4y2Ba学习者gydF4y2Ba
)가SVM을구현하는선형분류모델또는커널분류모델입니다。선형분류모델또는커널분류모델에대한사후확률을얻으려면이대신로지스틱회귀를구현하십시오。gydF4y2Ba
예:gydF4y2Ba“是的,没错gydF4y2Ba
데이터형:gydF4y2Ba逻辑gydF4y2Ba
学习者gydF4y2Ba
- - - - - -gydF4y2Ba이진학습기템플릿gydF4y2Ba“支持向量机”gydF4y2Ba
(디폴트 값) |gydF4y2Ba“判别”gydF4y2Ba
|gydF4y2Ba“内核”gydF4y2Ba
|gydF4y2Ba“资讯”gydF4y2Ba
|gydF4y2Ba“线性”gydF4y2Ba
|gydF4y2Ba“朴素贝叶斯”gydF4y2Ba
|gydF4y2Ba“树”gydF4y2Ba
|gydF4y2Ba템플릿객체gydF4y2Ba|gydF4y2Ba템플릿객체로구성된셀형벡터gydF4y2Ba이진 학습기 템플릿으로,gydF4y2Ba“学习者”gydF4y2Ba
와함께문자형벡터,弦형스칼라,템플릿객체,또는템플릿객체로구성된셀형벡터가쉼표로구분되어지정됩니다。특히,SVM과같은이진분류기와gydF4y2BaGentleBoostgydF4y2Ba
,gydF4y2BaLogitBoostgydF4y2Ba
,gydF4y2BaRobustBoostgydF4y2Ba
를 사용하는 앙상블을 지정하여 다중클래스 문제를 풀 수 있습니다. 그러나,gydF4y2BafitcecocgydF4y2Ba
는다중클래스모델도이진분류기로지원합니다。gydF4y2Ba
学习者gydF4y2Ba
가문자형벡터또는字符串형스칼라이면소프트웨어는지정된알고리즘의디폴트값을사용하여각각의이진학습기를훈련시킵니다。다음표에는사용가능한알고리즘이요약되어있습니다。gydF4y2Ba
값gydF4y2Ba | 설명gydF4y2Ba |
---|---|
“判别”gydF4y2Ba |
판별분석。디폴트옵션은gydF4y2BatemplateDiscriminantgydF4y2Ba 를 참조하십시오.gydF4y2Ba |
“内核”gydF4y2Ba |
커널 분류 모델. 디폴트 옵션은gydF4y2Ba模板核gydF4y2Ba 을참조하십시오。gydF4y2Ba |
“资讯”gydF4y2Ba |
k -최근접이웃。디폴트옵션은gydF4y2BatemplateKNNgydF4y2Ba 을참조하십시오。gydF4y2Ba |
“线性”gydF4y2Ba |
선형분류모델。디폴트옵션은gydF4y2BatemplateLineargydF4y2Ba 를 참조하십시오.gydF4y2Ba |
“朴素贝叶斯”gydF4y2Ba |
나이브베이즈(朴素贝叶斯)。디폴트옵션은gydF4y2BatemplateNaiveBayesgydF4y2Ba 를 참조하십시오.gydF4y2Ba |
“支持向量机”gydF4y2Ba |
支持向量机。디폴트옵션은gydF4y2BatemplateSVMgydF4y2Ba 을참조하십시오。gydF4y2Ba |
“树”gydF4y2Ba |
분류트리。디폴트옵션은gydF4y2BatemplateTreegydF4y2Ba 를 참조하십시오.gydF4y2Ba |
学习者gydF4y2Ba
가템플릿객체이면각각의이진학습기는저장된옵션을따라훈련합니다。다음을사용하여템플릿객체를생성할수있습니다。gydF4y2Ba
templateDiscriminantgydF4y2Ba
——판별분석에사용합니다。gydF4y2Ba
templateEnsemblegydF4y2Ba
——앙상블학습에사용합니다。최소한학습방법(gydF4y2Ba方法gydF4y2Ba
),학습기개수(gydF4y2BaNLearngydF4y2Ba
),학습기유형(gydF4y2Ba学习者gydF4y2Ba
)을지정해야합니다。이진학습에는gydF4y2BaAdaBoostM2gydF4y2Ba
앙상블방법을사용할수없습니다。gydF4y2Ba
模板核gydF4y2Ba
——커널분류에사용합니다。gydF4y2Ba
templateKNNgydF4y2Ba
- k최근접이웃에사용합니다。gydF4y2Ba
templateLineargydF4y2Ba
——선형분류에사용합니다。gydF4y2Ba
templateNaiveBayesgydF4y2Ba
- - - - - -나이브베이즈에사용합니다。gydF4y2Ba
templateSVMgydF4y2Ba
- - - - - - SVM에사용합니다。gydF4y2Ba
templateTreegydF4y2Ba
——분류트리에사용합니다。gydF4y2Ba
学习者gydF4y2Ba
가 템플릿 객체로 구성된 셀형 벡터이면 다음과 같습니다.gydF4y2Ba
셀j가이진학습기j(즉,코딩설계행렬의열j)에대응되고,셀형벡터의길이는L이어야합니다。L은코딩설계행렬에포함된열의개수입니다。자세한내용은gydF4y2Ba编码gydF4y2Ba
을참조하십시오。gydF4y2Ba
예측시내장손실함수중하나를사용하려면모든이진학습기가동일한범위의점수를반환해야합니다。예를들어디,폴트SVM이진학습기를디폴트나이브베이즈이진학습기와함께포함시킬수는없습니다。전자는(-∞∞)범위의점수를반환하고후자는사후확률을점수로반환합니다。또는,gydF4y2Ba预测gydF4y2Ba
및gydF4y2Ba损失gydF4y2Ba
와같은함수에대한함수핸들로사용자지정손실함수를제공해야합니다。gydF4y2Ba
선형분류모델학습기템플릿을다른템플릿과함께지정할수없습니다。gydF4y2Ba
마찬가지로,커널분류모델학습기템플릿을다른템플릿과함께지정할수없습니다。gydF4y2Ba
기본적으로,소프트웨어는디폴트SVM템플릿을사용하여학습기를훈련시킵니다。gydF4y2Ba
예:gydF4y2Ba“学习者”,“树”gydF4y2Ba
NumBinsgydF4y2Ba
- - - - - -gydF4y2Ba숫자형예측변수의本개수gydF4y2Ba[]gydF4y2Ba
(비어있음)gydF4y2Ba(디폴트 값) |gydF4y2Ba양의정수스칼라gydF4y2Ba숫자형예측변수의本개수로,gydF4y2Ba“NumBins”gydF4y2Ba
와함께양의정수스칼라가쉼표로구분되어지정됩니다。이인수는gydF4y2BafitcecocgydF4y2Ba
함수가트리학습기를사용하는경우에만유효합니다。즉,gydF4y2Ba“学习者”gydF4y2Ba
가gydF4y2Ba“树”gydF4y2Ba
이거나,gydF4y2BatemplateTreegydF4y2Ba
를사용하여만든템플릿객체이거나,약한트리학습기로gydF4y2BatemplateEnsemblegydF4y2Ba
을사용하여만든템플릿객체인경우에만유효합니다。gydF4y2Ba
“NumBins”gydF4y2Ba
값이비어있으면(디폴트값)gydF4y2BafitcecocgydF4y2Ba
함수는예측변수를비닝하지않습니다。gydF4y2Ba
“NumBins”gydF4y2Ba
값을양의정수스칼라(gydF4y2BanumBinsgydF4y2Ba
)로지정하면gydF4y2BafitcecocgydF4y2Ba
함수는모든숫자형예측변수를최대gydF4y2BanumBinsgydF4y2Ba
개의등확률本으로비닝한다음원래데이터가아닌本인덱스에서트리를성장시킵니다。gydF4y2Ba
예측변수의고유한값이gydF4y2BanumBinsgydF4y2Ba
개보다작은경우本의개수는gydF4y2BanumBinsgydF4y2Ba
개보다작을수있습니다。gydF4y2Ba
fitcecocgydF4y2Ba
함수는범주형예측변수를비닝하지않습니다。gydF4y2Ba
큰훈련데이터세트를사용할경우이비닝옵션은훈련속도를높이지만정확도가떨어질가능성이있습니다。먼저gydF4y2Ba“NumBins”,50岁gydF4y2Ba
을사용해본후에정확도와훈련속도에따라값을변경해볼수있습니다。gydF4y2Ba
훈련된모델은本경계값을gydF4y2BaBinEdgesgydF4y2Ba
속성에저장합니다。gydF4y2Ba
예:gydF4y2Ba“NumBins”,50岁gydF4y2Ba
데이터형:gydF4y2Ba单gydF4y2Ba
|gydF4y2Ba双重的gydF4y2Ba
NumConcurrentgydF4y2Ba
- - - - - -gydF4y2Ba동시에훈련되는이진학습기개수gydF4y2Ba1gydF4y2Ba
(디폴트 값) |gydF4y2Ba양의정수스칼라gydF4y2Ba동시에 훈련되는 이진 학습기 개수로,gydF4y2Ba“NumConcurrent”gydF4y2Ba
와함께양의정수스칼라가쉼표로구분되어지정됩니다。디폴트값은gydF4y2Ba1gydF4y2Ba
이며,이경우gydF4y2BafitcecocgydF4y2Ba
가이진학습기를순차적으로훈련시킵니다。gydF4y2Ba
참고gydF4y2Ba
이옵션은高형배열에gydF4y2BafitcecocgydF4y2Ba
를사용하는경우에만적용됩니다。자세한내용은gydF4y2Ba高형배열gydF4y2Ba항목을참조하십시오。gydF4y2Ba
데이터형:gydF4y2Ba单gydF4y2Ba
|gydF4y2Ba双重的gydF4y2Ba
ObservationsIngydF4y2Ba
- - - - - -gydF4y2Ba예측변수데이터관측차원gydF4y2Ba“行”gydF4y2Ba
(디폴트 값) |gydF4y2Ba“专栏”gydF4y2Ba
예측변수데이터관측차원으로,gydF4y2Ba“ObservationsIn”gydF4y2Ba
과함께gydF4y2Ba“专栏”gydF4y2Ba
또는gydF4y2Ba“行”gydF4y2Ba
가쉼표로구분되어지정됩니다。gydF4y2Ba
참고gydF4y2Ba
선형분류학습기의경우,관측값이열에대응되도록gydF4y2BaXgydF4y2Ba
의방향을지정하고gydF4y2Ba“ObservationsIn”、“列”gydF4y2Ba
를지정하면최적화-실행시간을상당히줄일수있습니다。gydF4y2Ba
다른모든학습기의경우관측값이행에대응되도록gydF4y2BaXgydF4y2Ba
의방향을지정하십시오。gydF4y2Ba
예:gydF4y2Ba“ObservationsIn”、“列”gydF4y2Ba
详细的gydF4y2Ba
- - - - - -gydF4y2Ba세부정보표시수준gydF4y2Ba0gydF4y2Ba
(디폴트 값) |gydF4y2Ba1gydF4y2Ba
|gydF4y2Ba2gydF4y2Ba
세부정보표시수준으로,gydF4y2Ba“详细”gydF4y2Ba
와함께gydF4y2Ba0gydF4y2Ba
,gydF4y2Ba1gydF4y2Ba
또는gydF4y2Ba2gydF4y2Ba
가쉼표로구분되어지정됩니다。gydF4y2Ba详细的gydF4y2Ba
는소프트웨어가명령창에표시하는이진학습기당진단정보의양을제어합니다。gydF4y2Ba
다음표에는사용가능한세부정보표시수준옵션이요약되어있습니다。gydF4y2Ba
값gydF4y2Ba | 설명gydF4y2Ba |
---|---|
0gydF4y2Ba |
소프트웨어가진단정보를표시하지않습니다。gydF4y2Ba |
1gydF4y2Ba |
소프트웨어가새로운이진학습기를훈련시킬때마다진단메시지를표시합니다。gydF4y2Ba |
2gydF4y2Ba |
소프트웨어가새로운이진학습기를훈련시킬때마다추가진단메시지를표시합니다。gydF4y2Ba |
각이진학습기는이이름——값쌍의인수와는독립적인고유한세부정보표시수준을갖습니다。이진학습기의세부정보표시수준을변경하려면템플릿객체를생성하고gydF4y2Ba“详细”gydF4y2Ba
이름——값쌍의인수를지정하십시오。그런다음,gydF4y2Ba“学习者”gydF4y2Ba
이름——값쌍의인수를사용하여템플릿객체를gydF4y2BafitcecocgydF4y2Ba
에전달합니다。gydF4y2Ba
예:gydF4y2Ba“详细”,1gydF4y2Ba
데이터형:gydF4y2Ba双重的gydF4y2Ba
|gydF4y2Ba单gydF4y2Ba
CrossValgydF4y2Ba
- - - - - -gydF4y2Ba교차검증된분류기를훈련시키는플래그gydF4y2Ba“关闭”gydF4y2Ba
(디폴트 값) |gydF4y2Ba“上”gydF4y2Ba
교차검증된분류기를훈련시키는플래그로,gydF4y2Ba“Crossval”gydF4y2Ba
과함께gydF4y2Ba“上”gydF4y2Ba
또는gydF4y2Ba“关闭”gydF4y2Ba
가쉼표로구분되어지정됩니다。gydF4y2Ba
“上”gydF4y2Ba
으로지정하면소프트웨어가10겹교차검증된분류기를훈련시킵니다。gydF4y2Ba
CVPartitiongydF4y2Ba
,gydF4y2Ba坚持gydF4y2Ba
,gydF4y2BaKFoldgydF4y2Ba
또는gydF4y2BaLeaveoutgydF4y2Ba
이름——값쌍의인수중하나를사용하여이교차검증설정을재정의할수있습니다。한번에하나의교차검증이름——값쌍의인수만사용하여교차검증된모델을생성할수있습니다。gydF4y2Ba
또는,gydF4y2BaMdlgydF4y2Ba
을gydF4y2BacrossvalgydF4y2Ba
로전달하여나중에교차검증을수행할수있습니다。gydF4y2Ba
예:gydF4y2Ba“Crossval”,“上”gydF4y2Ba
CVPartitiongydF4y2Ba
- - - - - -gydF4y2Ba교차검증분할gydF4y2Ba[]gydF4y2Ba
(디폴트 값) |gydF4y2BacvpartitiongydF4y2Ba
분할객체gydF4y2Ba교차 검증 분할로,gydF4y2BacvpartitiongydF4y2Ba
으로생성된gydF4y2BacvpartitiongydF4y2Ba
분할객체로지정됩니다。분할객체는교차검증의유형을지정하며훈련세트와검증세트의인덱싱도지정합니다。gydF4y2Ba
교차검증된모델을생성하려면다음4개의이름——값인수중하나만지정할수있습니다。gydF4y2BaCVPartitiongydF4y2Ba
,gydF4y2Ba坚持gydF4y2Ba
,gydF4y2BaKFoldgydF4y2Ba
,gydF4y2BaLeaveoutgydF4y2Ba
예:gydF4y2Ba本量利= cvpartition(500年,“KFold”,5)gydF4y2Ba
500개를사용하여관측값에대한5겹교차검증에사용할임의분할을생성한다고가정하겠습니다。그런다음,gydF4y2Ba“CVPartition”,本量利gydF4y2Ba
를사용하여교차검증된모델을지정할수있습니다。gydF4y2Ba
坚持gydF4y2Ba
- - - - - -gydF4y2Ba홀드아웃검증에사용할데이터의비율gydF4y2Ba홀드아웃 검증에 사용할 데이터의 비율로, 범위 (0,1) 내 스칼라 값으로 지정됩니다.gydF4y2Ba‘坚持’,pgydF4y2Ba
를 지정하는 경우 소프트웨어는 다음 단계를 완료합니다.gydF4y2Ba
데이터의gydF4y2Bap*100gydF4y2Ba
%를무작위로선택하여검증데이터용으로남겨두고나머지데이터를사용하여모델을훈련시킵니다。gydF4y2Ba
교차검증된모델의gydF4y2Ba训练有素的gydF4y2Ba
속성에훈련된간소모델을저장합니다。gydF4y2Ba
교차검증된모델을생성하려면다음4개의이름——값인수중하나만지정할수있습니다。gydF4y2BaCVPartitiongydF4y2Ba
,gydF4y2Ba坚持gydF4y2Ba
,gydF4y2BaKFoldgydF4y2Ba
,gydF4y2BaLeaveoutgydF4y2Ba
예:gydF4y2Ba“坚持”,0.1gydF4y2Ba
데이터형:gydF4y2Ba双重的gydF4y2Ba
|gydF4y2Ba单gydF4y2Ba
KFoldgydF4y2Ba
- - - - - -gydF4y2Ba겹의개수gydF4y2Ba10gydF4y2Ba
(디폴트 값) |gydF4y2Ba1보다큰양의정수값gydF4y2Ba교차검증된모델에사용할겹의개수로,1보다큰양의정수값으로지정됩니다。gydF4y2BaKFold, kgydF4y2Ba
를 지정하는 경우 소프트웨어는 다음 단계를 완료합니다.gydF4y2Ba
데이터를gydF4y2BakgydF4y2Ba
개세트로임의로분할합니다。gydF4y2Ba
각세트마다해당세트를검증데이터로남겨두고나머지gydF4y2BakgydF4y2Ba
– 1gydF4y2Ba개의세트를사용하여모델을훈련시킵니다。gydF4y2Ba
교차검증된모델의gydF4y2Ba训练有素的gydF4y2Ba
속성에gydF4y2BakgydF4y2Ba
×1셀형벡터로gydF4y2BakgydF4y2Ba
개의훈련된간소모델을저장합니다。gydF4y2Ba
교차검증된모델을생성하려면다음4개의이름——값인수중하나만지정할수있습니다。gydF4y2BaCVPartitiongydF4y2Ba
,gydF4y2Ba坚持gydF4y2Ba
,gydF4y2BaKFoldgydF4y2Ba
,gydF4y2BaLeaveoutgydF4y2Ba
예:gydF4y2Ba“KFold”,5gydF4y2Ba
데이터형:gydF4y2Ba单gydF4y2Ba
|gydF4y2Ba双重的gydF4y2Ba
LeaveoutgydF4y2Ba
- - - - - -gydF4y2Ba리브-원아웃(分析)교차검증플래그gydF4y2Ba“关闭”gydF4y2Ba
(디폴트 값) |gydF4y2Ba“上”gydF4y2Ba
리브-원아웃교차검증플래그로,gydF4y2Ba“Leaveout”gydF4y2Ba
과함께gydF4y2Ba“上”gydF4y2Ba
이나gydF4y2Ba“关闭”gydF4y2Ba
가쉼표로구분되어지정됩니다。gydF4y2Ba“Leaveout”,“上”gydF4y2Ba
을지정하는경우,n개관측값각각에대해(여기서n은gydF4y2Ba尺寸(Mdl.X, 1)gydF4y2Ba
임)소프트웨어가다음동작을수행합니다。gydF4y2Ba
관측값을검증데이터로남겨두고나머지n - 1개관측값을사용하여모델을훈련시킵니다。gydF4y2Ba
교차검증된모델의gydF4y2Ba训练有素的gydF4y2Ba
속성에n×1셀형벡터의셀로n개의훈련된간소모델을저장합니다。gydF4y2Ba
교차검증된모델을생성하려면다음4개의옵션중하나만을사용할수있습니다。gydF4y2BaCVPartitiongydF4y2Ba
,gydF4y2Ba坚持gydF4y2Ba
,gydF4y2BaKFoldgydF4y2Ba
또는gydF4y2BaLeaveoutgydF4y2Ba
.gydF4y2Ba
참고gydF4y2Ba
리브-원아웃은선형분류모델학습기또는커널분류모델학습기로구성된ECOC모델을교차검증하는데는권장되지않습니다。gydF4y2Ba
예:gydF4y2Ba“Leaveout”,“上”gydF4y2Ba
分类预测因子gydF4y2Ba
- - - - - -gydF4y2Ba범주형예측변수목록gydF4y2Ba“所有”gydF4y2Ba
범주형예측변수목록으로,다음표에있는값중하나로지정됩니다。gydF4y2Ba
값gydF4y2Ba | 설명gydF4y2Ba |
---|---|
양의정수로구성된벡터gydF4y2Ba | 벡터의각요소는범주형변수를포함하는예측변수데이터의열에대응되는인덱스값입니다。과인덱스값은1gydF4y2Ba
|
논리형 벡터gydF4y2Ba | 요소의값이gydF4y2Ba |
문자형행렬gydF4y2Ba | 행렬의각행은예측변수의이름입니다。이름은gydF4y2BaPredictorNamesgydF4y2Ba 의요소와일치해야합니다。문자형행렬의각행의길이가같게되도록이름뒤에추가로공백을채웁니다。gydF4y2Ba |
字符串형배열또는문자형벡터로구성된셀형배열gydF4y2Ba | 배열의각요소는예측변수의이름입니다。이름은gydF4y2BaPredictorNamesgydF4y2Ba 의요소와일치해야합니다。gydF4y2Ba |
“所有”gydF4y2Ba |
모든예측변수가범주형변수입니다。gydF4y2Ba |
다음과같은경우gydF4y2Ba“CategoricalPredictors”gydF4y2Ba
를지정하는것이적합합니다。gydF4y2Ba
하나이상의예측변수가범주형변수이고모든이진학습기가분류트리,나이브베이즈학습기,支持向量机,선형학습기,커널학습기또는분류트리의앙상블입니다。gydF4y2Ba
모든 예측 변수가 범주형 변수이고 하나 이상의 이진 학습기가 kNN입니다.gydF4y2Ba
그밖의학습기에gydF4y2Ba“CategoricalPredictors”gydF4y2Ba
를지정하면이진학습기를훈련시킬수없다는내용의경고가표시됩니다。예를들어,범주형예측변수를사용하여판별분석분류기를훈련시킬수없습니다。gydF4y2Ba
각학습기는해당학습기에서사용하는피팅함수와동일한방식으로범주형예측변수를식별하고처리합니다。커널학습기는gydF4y2BafitckernelgydF4y2Ba
의gydF4y2Ba“CategoricalPredictors”gydF4y2Ba
k -최근접학습기는gydF4y2BafitcknngydF4y2Ba
의gydF4y2Ba“CategoricalPredictors”gydF4y2Ba
, 선형 학습기는gydF4y2BafitclineargydF4y2Ba
의gydF4y2Ba“CategoricalPredictors”gydF4y2Ba
,나이브베이즈학습기는gydF4y2BafitcnbgydF4y2Ba
의gydF4y2Ba“CategoricalPredictors”gydF4y2Ba
SVM학습기는gydF4y2BafitcsvmgydF4y2Ba
의gydF4y2Ba“CategoricalPredictors”gydF4y2Ba
, 트리 학습기는gydF4y2BafitctreegydF4y2Ba
의gydF4y2Ba“CategoricalPredictors”gydF4y2Ba
를 참조하십시오.gydF4y2Ba
예:gydF4y2Ba“CategoricalPredictors”、“所有”gydF4y2Ba
데이터형:gydF4y2Ba单gydF4y2Ba
|gydF4y2Ba双重的gydF4y2Ba
|gydF4y2Ba逻辑gydF4y2Ba
|gydF4y2Ba字符gydF4y2Ba
|gydF4y2Ba字符串gydF4y2Ba
|gydF4y2Ba细胞gydF4y2Ba
一会gydF4y2Ba
- - - - - -gydF4y2Ba훈련에사용할클래스의이름gydF4y2Ba훈련에사용할클래스의이름으로,直言형배열,문자형배열,字符串형배열,논리형벡터또는숫자형벡터,문자형벡터로구성된셀형배열로지정됩니다。gydF4y2Ba一会gydF4y2Ba
는gydF4y2Ba资源描述gydF4y2Ba
의응답변수또는gydF4y2BaYgydF4y2Ba
와같은데이터형이어야합니다。gydF4y2Ba
一会gydF4y2Ba
가문자형배열인경우,각요소는배열의각행에대응되어야합니다。gydF4y2Ba
一会gydF4y2Ba
를사용하여다음을수행할수있습니다。gydF4y2Ba
훈련중의클래스순서를지정합니다。gydF4y2Ba
입력인수차원또는출력인수차원의순서를지정합니다。이순서는클래스순서와일치합니다。예를들어,gydF4y2Ba一会gydF4y2Ba
를사용하여gydF4y2Ba成本gydF4y2Ba
차원의순서나gydF4y2Ba预测gydF4y2Ba
로반환되는분류점수의열순서를지정할수있습니다。gydF4y2Ba
훈련에 사용할 클래스의 일부를 선택합니다. 예를 들어,gydF4y2BaYgydF4y2Ba
에포함된모든고유한클래스이름의집합이gydF4y2Ba{' a ', ' b ', ' c '}gydF4y2Ba
라고가정해보겠습니다。클래스gydF4y2Ba“一个”gydF4y2Ba
와gydF4y2Ba“c”gydF4y2Ba
의관측값만사용하여모델을훈련시키려면gydF4y2Ba“类名”,{' a ', ' c '}gydF4y2Ba
를지정하십시오。gydF4y2Ba
一会gydF4y2Ba
의디폴트값은gydF4y2Ba资源描述gydF4y2Ba
의응답변수또는gydF4y2BaYgydF4y2Ba
에 포함된 모든 고유한 클래스 이름의 집합입니다.gydF4y2Ba
예:gydF4y2Ba“类名”,{' b ', ' g '}gydF4y2Ba
데이터형:gydF4y2Ba分类gydF4y2Ba
|gydF4y2Ba字符gydF4y2Ba
|gydF4y2Ba字符串gydF4y2Ba
|gydF4y2Ba逻辑gydF4y2Ba
|gydF4y2Ba单gydF4y2Ba
|gydF4y2Ba双重的gydF4y2Ba
|gydF4y2Ba细胞gydF4y2Ba
成本gydF4y2Ba
- - - - - -gydF4y2Ba오분류비용gydF4y2Ba오분류비용으로,gydF4y2Ba“成本”gydF4y2Ba
와함께정사각행렬또는구조체가쉼표로구분되어지정됩니다。다음을참고하십시오。gydF4y2Ba
정사각행렬gydF4y2Ba成本gydF4y2Ba
를지정하는경우,gydF4y2Ba成本(i, j)gydF4y2Ba
는실제클래스가gydF4y2Ba我gydF4y2Ba
인한점을클래스gydF4y2BajgydF4y2Ba
로 분류하는 비용입니다. 즉, 행은 실제 클래스에 대응되고, 열은 예측 클래스에 대응됩니다.gydF4y2Ba成本gydF4y2Ba
의대응행과대응열에대한클래스순서를지정하려면gydF4y2Ba一会gydF4y2Ba
이름-값 쌍의 인수도 지정하십시오.gydF4y2Ba
구조체gydF4y2Ba年代gydF4y2Ba
를지정하는경우이,구조체는다음과같은두개의필드를가져야합니다。gydF4y2Ba
클래스이름을gydF4y2BaYgydF4y2Ba
와같은데이터형의변수로포함하는gydF4y2BaS.ClassNamesgydF4y2Ba
S.ClassNamesgydF4y2Ba
에서와같은순서로행과열을갖는비용행렬을포함하는gydF4y2Ba美国分类成本gydF4y2Ba
디폴트값은gydF4y2Ba(gydF4y2Ba
입니다。여기서gydF4y2BaKgydF4y2Ba
) - (gydF4y2BaKgydF4y2Ba
)gydF4y2BaKgydF4y2Ba
는고유한클래스의개수입니다。gydF4y2Ba
예:gydF4y2Ba'Cost',[0 1 2;1 0 2;2 2 0)gydF4y2Ba
데이터형:gydF4y2Ba双重的gydF4y2Ba
|gydF4y2Ba单gydF4y2Ba
|gydF4y2Ba结构体gydF4y2Ba
选项gydF4y2Ba
- - - - - -gydF4y2Ba병렬 연산 옵션gydF4y2Ba[]gydF4y2Ba
(디폴트 값) |gydF4y2Ba斯塔塞特gydF4y2Ba
에서반환되는구조체형배열gydF4y2Ba병렬연산옵션으로,gydF4y2Ba“选项”gydF4y2Ba
와함께gydF4y2Ba斯塔塞特gydF4y2Ba
에서반환되는구조체형배열이쉼표로구분되어지정됩니다。이옵션을사용하려면并行计算工具箱™가필요합니다。gydF4y2BafitcecocgydF4y2Ba
함수는gydF4y2Ba“流”gydF4y2Ba
필드,gydF4y2Ba“UseParallel”gydF4y2Ba
필드,gydF4y2Ba“UseSubtreams”gydF4y2Ba
필드를사용합니다。gydF4y2Ba
다음표에는사용가능한옵션이요약되어있습니다。gydF4y2Ba
옵션gydF4y2Ba | 설명gydF4y2Ba |
---|---|
“流”gydF4y2Ba |
이경우,병렬풀과크기가같은셀형배열을사용하십시오。병렬풀이열려있지않으면소프트웨어가병렬풀(기본설정에따름)을열려고하고gydF4y2Ba |
“UseParallel”gydF4y2Ba |
并行计算工具箱가있는경우gydF4y2Ba 이진학습기에결정트리를사용할때는gydF4y2Ba |
“UseSubstreams”gydF4y2Ba |
“流”gydF4y2Ba 로지정된스트림을사용하여병렬로계산하려면gydF4y2Ba真正的gydF4y2Ba 를 설정하십시오. 디폴트 값은gydF4y2Ba假gydF4y2Ba 입니다。예를들어,gydF4y2Ba“mlfg6331_64”gydF4y2Ba 또는gydF4y2Ba“mrg32k3a”gydF4y2Ba 와같은서브스트림을허용하는유형으로gydF4y2Ba溪流gydF4y2Ba 를 설정하십시오.gydF4y2Ba |
더욱예측가능한결과를얻으려면gydF4y2BaparpoolgydF4y2Ba
(并行计算工具箱)gydF4y2Ba을사용하고gydF4y2BafitcecocgydF4y2Ba
함수를사용하여병렬계산을불러오기전에병렬풀을명시적으로생성하는것이좋습니다。gydF4y2Ba
예:gydF4y2Ba“选项”,statset('UseParallel',true)gydF4y2Ba
데이터형:gydF4y2Ba结构体gydF4y2Ba
PredictorNamesgydF4y2Ba
- - - - - -gydF4y2Ba예측변수이름gydF4y2Ba예측 변수 이름으로, 고유한 이름으로 구성된 一串형 배열 또는 고유한 문자형 벡터로 구성된 셀형 배열로 지정됩니다.gydF4y2BaPredictorNamesgydF4y2Ba
의기능은훈련데이터를어떤방식으로제공하느냐에따라달라집니다。gydF4y2Ba
XgydF4y2Ba
와gydF4y2BaYgydF4y2Ba
를제공하는경우,gydF4y2BaPredictorNamesgydF4y2Ba
를사용하여gydF4y2BaXgydF4y2Ba
의예측변수에이름을할당할수있습니다。gydF4y2Ba
PredictorNamesgydF4y2Ba
의이름의순서는gydF4y2BaXgydF4y2Ba
의열순서와일치해야합니다。즉,gydF4y2BaPredictorNames {1}gydF4y2Ba
은gydF4y2BaX (: 1)gydF4y2Ba
의이름이고,gydF4y2BaPredictorNames {2}gydF4y2Ba
는gydF4y2BaX (:, 2)gydF4y2Ba
의이름이되는식입니다。또한,gydF4y2Ba大小(X, 2)gydF4y2Ba
와gydF4y2Ba元素个数(PredictorNames)gydF4y2Ba
는 같아야 합니다.gydF4y2Ba
기본적으로gydF4y2BaPredictorNamesgydF4y2Ba
는gydF4y2Ba{x1, x2,…}gydF4y2Ba
입니다。gydF4y2Ba
资源描述gydF4y2Ba
을제공하는경우,gydF4y2BaPredictorNamesgydF4y2Ba
를사용하여훈련에사용할예측변수를선택할수있습니다。즉,gydF4y2BafitcecocgydF4y2Ba
함수는gydF4y2BaPredictorNamesgydF4y2Ba
의예측변수와이에대한응답변수만을훈련중에사용합니다。gydF4y2Ba
PredictorNamesgydF4y2Ba
는gydF4y2BaTbl.Properties.VariableNamesgydF4y2Ba
의부분집합이어야하므로응답변수의이름은포함할수없습니다。gydF4y2Ba
기본적으로,gydF4y2BaPredictorNamesgydF4y2Ba
는모든예측변수의이름을포함합니다。gydF4y2Ba
“PredictorNames”gydF4y2Ba
와gydF4y2Ba公式gydF4y2Ba
중하나만사용하여훈련에사용할예측변수를지정하는것이좋습니다。gydF4y2Ba
예:gydF4y2BaPredictorNames,{‘SepalLength’,‘SepalWidth’,‘PetalLength’,‘PetalWidth}gydF4y2Ba
데이터형:gydF4y2Ba字符串gydF4y2Ba
|gydF4y2Ba细胞gydF4y2Ba
之前gydF4y2Ba
- - - - - -gydF4y2Ba사전확률gydF4y2Ba“经验”gydF4y2Ba
(디폴트 값) |gydF4y2Ba“统一”gydF4y2Ba
|gydF4y2Ba숫자형벡터gydF4y2Ba|gydF4y2Ba구조체형배열gydF4y2Ba각클래스의사전확률로,gydF4y2Ba“之前”gydF4y2Ba
와함께다음표에나와있는값이쉼표로구분되어지정됩니다。gydF4y2Ba
값gydF4y2Ba | 설명gydF4y2Ba |
---|---|
“经验”gydF4y2Ba |
클래스사전확률은gydF4y2BaYgydF4y2Ba 의클래스상대빈도입니다。gydF4y2Ba |
“统一”gydF4y2Ba |
모든클래스사전확률은1 / K와같습니다。여기K서는클래스개수입니다。gydF4y2Ba |
숫자형벡터gydF4y2Ba | 각요소는클래스사전확률입니다。gydF4y2BaMdlgydF4y2Ba .ClassNamesgydF4y2Ba 에 따라 요소의 순서를 지정하거나gydF4y2Ba一会gydF4y2Ba 이름——값쌍의인수를사용하여순서를지정합니다。소프트웨어는합이gydF4y2Ba1gydF4y2Ba 이되도록요소를정규화합니다。gydF4y2Ba |
구조체gydF4y2Ba | 다음과같은두개의필드를갖는구조체gydF4y2Ba
|
소프트웨어가클래스사전확률을통합하는방법에대한자세한내용은gydF4y2Ba사전 확률과 비용gydF4y2Ba항목을참조하십시오。gydF4y2Ba
예:gydF4y2Ba结构(“类名”,{{setosa,杂色的,‘virginica}}, ClassProbs, 1:3)gydF4y2Ba
데이터형:gydF4y2Ba单gydF4y2Ba
|gydF4y2Ba双重的gydF4y2Ba
|gydF4y2Ba字符gydF4y2Ba
|gydF4y2Ba字符串gydF4y2Ba
|gydF4y2Ba结构体gydF4y2Ba
ResponseNamegydF4y2Ba
- - - - - -gydF4y2Ba응답변수이름gydF4y2Ba“Y”gydF4y2Ba
(디폴트 값) |gydF4y2Ba문자형벡터gydF4y2Ba|gydF4y2Ba字符串형스칼라gydF4y2Ba응답 변수 이름으로, 문자형 벡터 또는 一串형 스칼라로 지정됩니다.gydF4y2Ba
YgydF4y2Ba
를제공하는경우,gydF4y2Ba“ResponseName”gydF4y2Ba
을사용하여응답변수의이름을지정할수있습니다。gydF4y2Ba
ResponseVarNamegydF4y2Ba
또는gydF4y2Ba公式gydF4y2Ba
를 제공하는 경우에는gydF4y2Ba“ResponseName”gydF4y2Ba
을사용할수없습니다。gydF4y2Ba
예:gydF4y2Ba“ResponseName”、“响应”gydF4y2Ba
데이터형:gydF4y2Ba字符gydF4y2Ba
|gydF4y2Ba字符串gydF4y2Ba
ScoreTransformgydF4y2Ba
- - - - - -gydF4y2Ba점수변환방식gydF4y2Ba“没有”gydF4y2Ba
(디폴트 값) |gydF4y2Ba“doublelogit”gydF4y2Ba
|gydF4y2Ba“invlogit”gydF4y2Ba
|gydF4y2Ba“ismax”gydF4y2Ba
|gydF4y2Ba分对数的gydF4y2Ba
|gydF4y2Ba함수핸들gydF4y2Ba|……gydF4y2Ba점수변환방식으로,문자형벡터,字符串형스칼라또는함수핸들로지정됩니다。gydF4y2Ba
다음표에는사용가능한문자형벡터와字符串형스칼라가요약되어있습니다。gydF4y2Ba
값gydF4y2Ba | 설명gydF4y2Ba |
---|---|
“doublelogit”gydF4y2Ba |
1 / (1 + egydF4y2Ba2 xgydF4y2Ba)gydF4y2Ba |
“invlogit”gydF4y2Ba |
Log (x / (1 - x))gydF4y2Ba |
“ismax”gydF4y2Ba |
최대점수를갖는클래스의점수를1로설정하고,다른모든클래스의점수를0으로설정합니다。gydF4y2Ba |
分对数的gydF4y2Ba |
1 / (1 + egydF4y2Ba- xgydF4y2Ba)gydF4y2Ba |
“没有”gydF4y2Ba 또는gydF4y2Ba“身份”gydF4y2Ba |
x(변환없음)gydF4y2Ba |
“标志”gydF4y2Ba |
x < 0의경우1gydF4y2Ba x=0의 경우 0gydF4y2Ba x > 0의경우1gydF4y2Ba |
“对称”gydF4y2Ba |
2 x - 1gydF4y2Ba |
“symmetricismax”gydF4y2Ba |
최대점수를갖는클래스의점수를1로설정하고,다른모든클래스의점수를1로설정합니다。gydF4y2Ba |
“symmetriclogit”gydF4y2Ba |
2 / (1 + egydF4y2Ba- xgydF4y2Ba) - 1gydF4y2Ba |
MATLAB함수나사용자가직접정의하는함수의경우,이에대한함수핸들을점수변환에사용하십시오。함수핸들은행렬(원래점수)을받아동일한크기의행렬(변환된점수)을반환합니다。gydF4y2Ba
예:gydF4y2Ba“ScoreTransform”、“分对数的gydF4y2Ba
데이터형:gydF4y2Ba字符gydF4y2Ba
|gydF4y2Ba字符串gydF4y2Ba
|gydF4y2Bafunction_handlegydF4y2Ba
权重gydF4y2Ba
- - - - - -gydF4y2Ba관측값가중치gydF4y2Ba资源描述gydF4y2Ba
에포함된변수이름gydF4y2Ba관측값가중치로,gydF4y2Ba“重量”gydF4y2Ba
와함께양수값으로구성된숫자형벡터나gydF4y2Ba资源描述gydF4y2Ba
에포함된변수의이름이쉼표로구분되어지정됩니다。소프트웨어는gydF4y2BaXgydF4y2Ba
또는gydF4y2Ba资源描述gydF4y2Ba
의각행에있는관측값에이에대응하는gydF4y2Ba权重gydF4y2Ba
의값을함께사용하여가중치를적용합니다。gydF4y2Ba权重gydF4y2Ba
의 크기는gydF4y2BaXgydF4y2Ba
또는gydF4y2Ba资源描述gydF4y2Ba
의행개수와일치해야합니다。gydF4y2Ba
입력데이터를테이블gydF4y2Ba资源描述gydF4y2Ba
로지정하는경우,gydF4y2Ba权重gydF4y2Ba
는gydF4y2Ba资源描述gydF4y2Ba
에서숫자형벡터를포함하는변수의이름일수있습니다。이경우,gydF4y2Ba权重gydF4y2Ba
를문자형벡터또는字符串형스칼라로지정해야합니다。,예를들어가중벡터gydF4y2BaWgydF4y2Ba
가gydF4y2Ba资源描述。WgydF4y2Ba
로저장된경우,이를gydF4y2Ba' W 'gydF4y2Ba
로지정하십시오。그렇지않은경우,소프트웨어는모델을훈련시킬때gydF4y2BaWgydF4y2Ba
를포함한gydF4y2Ba资源描述gydF4y2Ba
의모든열을예측변수또는응답변수로처리합니다。gydF4y2Ba
소프트웨어는gydF4y2Ba权重gydF4y2Ba
의 총합이 각 클래스의 사전 확률의 값이 되도록 정규화합니다.gydF4y2Ba
기본적으로,gydF4y2Ba权重gydF4y2Ba
는gydF4y2Ba(gydF4y2Ba
입니다。여기서gydF4y2BangydF4y2Ba
,1)gydF4y2BangydF4y2Ba
은gydF4y2BaXgydF4y2Ba
또는gydF4y2Ba资源描述gydF4y2Ba
에포함된관측값개수입니다。gydF4y2Ba
데이터형:gydF4y2Ba双重的gydF4y2Ba
|gydF4y2Ba单gydF4y2Ba
|gydF4y2Ba字符gydF4y2Ba
|gydF4y2Ba字符串gydF4y2Ba
OptimizeHyperparametersgydF4y2Ba
- - - - - -gydF4y2Ba최적화할모수gydF4y2Ba“没有”gydF4y2Ba
(디폴트 값) |gydF4y2Ba“汽车”gydF4y2Ba
|gydF4y2Ba“所有”gydF4y2Ba
|gydF4y2Ba적합한 모수 이름으로 구성된 一串형 배열 또는 셀형 배열gydF4y2Ba|gydF4y2BaoptimizableVariablegydF4y2Ba
객체로구성된벡터gydF4y2Ba최적화할모수로,gydF4y2Ba“OptimizeHyperparameters”gydF4y2Ba
와함께다음값중하나가쉼표로구분되어지정됩니다。gydF4y2Ba
“没有”gydF4y2Ba
(최적화하지않습니다。gydF4y2Ba
“汽车”gydF4y2Ba
- - - - - -지정된gydF4y2Ba学习者gydF4y2Ba
의디폴트모수와함께gydF4y2Ba{'Coding'}gydF4y2Ba
을사용합니다。gydF4y2Ba
学习者gydF4y2Ba
=gydF4y2Ba“支持向量机”gydF4y2Ba
(디폴트 값) —gydF4y2Ba{“BoxConstraint”、“KernelScale”}gydF4y2Ba
学习者gydF4y2Ba
=gydF4y2Ba“判别”gydF4y2Ba
- - - - - -gydF4y2Ba{“三角洲”,“伽马”}gydF4y2Ba
学习者gydF4y2Ba
=gydF4y2Ba“内核”gydF4y2Ba
- - - - - -gydF4y2Ba{“KernelScale”、“λ”}gydF4y2Ba
学习者gydF4y2Ba
=gydF4y2Ba“资讯”gydF4y2Ba
- - - - - -gydF4y2Ba{'Distance','numnighbors'}gydF4y2Ba
学习者gydF4y2Ba
=gydF4y2Ba“线性”gydF4y2Ba
- - - - - -gydF4y2Ba{'Lambda','Learner'}gydF4y2Ba
学习者gydF4y2Ba
=gydF4y2Ba“朴素贝叶斯”gydF4y2Ba
- - - - - -gydF4y2Ba{“DistributionNames”、“宽度”}gydF4y2Ba
学习者gydF4y2Ba
=gydF4y2Ba“树”gydF4y2Ba
- - - - - -gydF4y2Ba{' MinLeafSize '}gydF4y2Ba
“所有”gydF4y2Ba
——모든적합한모수를최적화합니다。gydF4y2Ba
적합한 모수 이름으로 구성된 一串형 배열 또는 셀형 배열gydF4y2Ba
optimizableVariablegydF4y2Ba
객체로구성된벡터。일반적으로gydF4y2BahyperparametersgydF4y2Ba
의출력값임gydF4y2Ba
최적화는모수를변경하여gydF4y2BafitcecocgydF4y2Ba
에대한교차검증손실(오차)을최소화하려고합니다。이와는다른맥락의교차검증손실에대한자세한내용은gydF4y2Ba分类损失gydF4y2Ba항목을참조하십시오。교차검증유형과최적화의기타측면을제어하려면gydF4y2Ba超参数优化选项gydF4y2Ba
이름——값쌍을사용하십시오。gydF4y2Ba
참고gydF4y2Ba
“OptimizeHyperparameters”gydF4y2Ba
값은다른이름——값쌍의인수를사용하여설정하는모든값을재정의합니다。예를들어,gydF4y2Ba“OptimizeHyperparameters”gydF4y2Ba
를gydF4y2Ba“汽车”gydF4y2Ba
로설정하면gydF4y2Ba“汽车”gydF4y2Ba
값이 적용됩니다.gydF4y2Ba
fitcecocgydF4y2Ba
에대한적합한모수는다음과같습니다。gydF4y2Ba
编码gydF4y2Ba
- - - - - -gydF4y2BafitcecocgydF4y2Ba
가gydF4y2Ba“onevsall”gydF4y2Ba
및gydF4y2Ba“onevsone”gydF4y2Ba
중에서탐색을수행합니다。gydF4y2Ba
아래표에서각gydF4y2Ba学习者gydF4y2Ba
마다지정된적합한하이퍼파라미터。gydF4y2Ba
학습기gydF4y2Ba | 적합한하이퍼파라미터gydF4y2Ba (굵게표시된항목이디폴트값임)gydF4y2Ba |
디폴트 범위gydF4y2Ba |
---|---|---|
“判别”gydF4y2Ba |
δgydF4y2Ba |
(1 e-6, 1 e3)gydF4y2Ba 범위에서로그스케일링된값gydF4y2Ba |
DiscrimTypegydF4y2Ba |
“线性”gydF4y2Ba ,gydF4y2Ba“二次”gydF4y2Ba ,gydF4y2Ba“对角线性”gydF4y2Ba ,gydF4y2Ba“diagQuadratic”gydF4y2Ba ,gydF4y2Ba“pseudoLinear”gydF4y2Ba ,gydF4y2Ba“pseudoQuadratic”gydF4y2Ba |
|
γgydF4y2Ba |
[0, 1]gydF4y2Ba 의실수값gydF4y2Ba |
|
“内核”gydF4y2Ba |
λgydF4y2Ba |
(1 e - 3 / NumObservations, e3 / NumObservations]gydF4y2Ba 범위에서로그스케일링된양수값gydF4y2Ba |
KernelScalegydF4y2Ba |
(1 e - 3, 1 e3)gydF4y2Ba 범위에서로그스케일링된양수값gydF4y2Ba |
|
学习者gydF4y2Ba |
“支持向量机”gydF4y2Ba 과gydF4y2Ba“物流”gydF4y2Ba |
|
NumExpansionDimensionsgydF4y2Ba |
(100、10000)gydF4y2Ba 범위에서로그스케일링된정수gydF4y2Ba |
|
“资讯”gydF4y2Ba |
距离gydF4y2Ba |
“cityblock”gydF4y2Ba ,gydF4y2Ba“chebychev”gydF4y2Ba ,gydF4y2Ba“相关”gydF4y2Ba ,gydF4y2Ba“余弦”gydF4y2Ba ,gydF4y2Ba“欧几里得”gydF4y2Ba ,gydF4y2Ba“汉明”gydF4y2Ba ,gydF4y2Ba“jaccard”gydF4y2Ba ,gydF4y2Ba“mahalanobis”gydF4y2Ba ,gydF4y2Ba闵可夫斯基的gydF4y2Ba ,gydF4y2Ba“seuclidean”gydF4y2Ba ,gydF4y2Ba“枪兵”gydF4y2Ba |
DistanceWeightgydF4y2Ba |
“平等”gydF4y2Ba ,gydF4y2Ba“逆”gydF4y2Ba ,gydF4y2Ba“squaredinverse”gydF4y2Ba |
|
指数gydF4y2Ba |
(0.5, 3)gydF4y2Ba 의양수값gydF4y2Ba |
|
NumNeighborsgydF4y2Ba |
[1,马克斯(2轮(NumObservations / 2)))gydF4y2Ba 범위에서로그스케일링된양의정수값gydF4y2Ba |
|
标准化gydF4y2Ba |
“真正的”gydF4y2Ba 와gydF4y2Ba“假”gydF4y2Ba |
|
“线性”gydF4y2Ba |
λgydF4y2Ba |
[1 e-5 / NumObservations 1 e5 / NumObservations]gydF4y2Ba 범위에서로그스케일링된양수값gydF4y2Ba |
学习者gydF4y2Ba |
“支持向量机”gydF4y2Ba 과gydF4y2Ba“物流”gydF4y2Ba |
|
正则化gydF4y2Ba |
“岭”gydF4y2Ba 와gydF4y2Ba“套索”gydF4y2Ba |
|
“朴素贝叶斯”gydF4y2Ba |
DistributionNamesgydF4y2Ba |
“正常”gydF4y2Ba 과gydF4y2Ba“内核”gydF4y2Ba |
宽度gydF4y2Ba |
[MinPredictorDiff / 4,马克斯(MaxPredictorRange MinPredictorDiff)]gydF4y2Ba 범위에서로그스케일링된양수값gydF4y2Ba |
|
内核gydF4y2Ba |
“正常”gydF4y2Ba ,gydF4y2Ba“盒子”gydF4y2Ba ,gydF4y2Ba“epanechnikov”gydF4y2Ba ,gydF4y2Ba“三角形”gydF4y2Ba |
|
“支持向量机”gydF4y2Ba |
BoxConstraintgydF4y2Ba |
(1 e - 3, 1 e3)gydF4y2Ba 범위에서로그스케일링된양수값gydF4y2Ba |
KernelScalegydF4y2Ba |
(1 e - 3, 1 e3)gydF4y2Ba 범위에서로그스케일링된양수값gydF4y2Ba |
|
KernelFunctiongydF4y2Ba |
“高斯”gydF4y2Ba ,gydF4y2Ba“线性”gydF4y2Ba ,gydF4y2Ba多项式的gydF4y2Ba |
|
多项式序gydF4y2Ba |
(2、4)gydF4y2Ba 범위의정수gydF4y2Ba |
|
标准化gydF4y2Ba |
“真正的”gydF4y2Ba 와gydF4y2Ba“假”gydF4y2Ba |
|
“树”gydF4y2Ba |
MaxNumSplitsgydF4y2Ba |
NumObservations-1[1,马克斯(2))gydF4y2Ba 범위에서로그스케일링된정수gydF4y2Ba |
MinLeafSizegydF4y2Ba |
[1,马克斯(2楼(NumObservations / 2)))gydF4y2Ba 범위에서로그스케일링된정수gydF4y2Ba |
|
NumVariablesToSamplegydF4y2Ba |
NumPredictors[1,马克斯(2))gydF4y2Ba 범위의정수gydF4y2Ba |
|
SplitCriteriongydF4y2Ba |
gdi的gydF4y2Ba ,gydF4y2Ba“异常”gydF4y2Ba ,gydF4y2Ba“两个”gydF4y2Ba |
또는,다음과같이선택한gydF4y2Ba学习者gydF4y2Ba
와함께gydF4y2BahyperparametersgydF4y2Ba
를사용할수있습니다。gydF4y2Ba
负载gydF4y2BafisheririsgydF4y2Ba超参数需要数据和学习者gydF4y2Baparams = hyperparameters (gydF4y2Ba“fitcecoc”gydF4y2Ba,meas,物种,gydF4y2Ba“支持向量机”gydF4y2Ba);gydF4y2Ba
적합한하이퍼파라미터와디폴트하이퍼파라미터를보려면gydF4y2Ba参数个数gydF4y2Ba
를살펴보십시오。gydF4y2Ba
디폴트가아닌값을가지는gydF4y2BaoptimizableVariablegydF4y2Ba
객체로구성된벡터를전달하여디폴트가아닌모수를설정합니다。예를들면다음과같습니다。gydF4y2Ba
负载gydF4y2BafisheririsgydF4y2Baparams = hyperparameters (gydF4y2Ba“fitcecoc”gydF4y2Ba,meas,物种,gydF4y2Ba“支持向量机”gydF4y2Ba);参数(2)。范围=(1的军医,1 e6);gydF4y2Ba
参数个数gydF4y2Ba
를gydF4y2BaOptimizeHyperparametersgydF4y2Ba
의값으로전달합니다。gydF4y2Ba
기본적으로,반복표시가명령줄에표시되고,최적화에지정된하이퍼파라미터개수에따라플롯이표시됩니다。최적화와플롯에대해목적함수는회귀의경우gydF4y2BaLog(1 +交叉验证损失)gydF4y2Ba분이고,류의경우오분류율입니다。반복표시를제어하려면gydF4y2Ba“HyperparameterOptimizationOptions”gydF4y2Ba
이름——값쌍의인수에대한gydF4y2Ba详细的gydF4y2Ba
필드를설정하십시오。플롯을제어하려면gydF4y2Ba“HyperparameterOptimizationOptions”gydF4y2Ba
이름——값쌍의인수에대한gydF4y2BaShowPlotsgydF4y2Ba
필드를설정하십시오。gydF4y2Ba
예제는gydF4y2Ba经济合作분류기 최적화하기gydF4y2Ba항목을참조하십시오。gydF4y2Ba
예:gydF4y2Ba“汽车”gydF4y2Ba
超参数优化选项gydF4y2Ba
- - - - - -gydF4y2Ba최적화에사용할옵션gydF4y2Ba최적화에사용할옵션으로,gydF4y2Ba“HyperparameterOptimizationOptions”gydF4y2Ba
와함께구조체가쉼표로구분되어지정됩니다。이인수는gydF4y2BaOptimizeHyperparametersgydF4y2Ba
이름——값쌍의인수의효과를수정합니다。이구조체에포함된모든필드는선택사항입니다。gydF4y2Ba
필드이름gydF4y2Ba | 값gydF4y2Ba | 디폴트 값gydF4y2Ba |
---|---|---|
优化器gydF4y2Ba |
|
“bayesopt”gydF4y2Ba |
AcquisitionFunctionNamegydF4y2Ba |
최적화는목적함수의런타임에종속적이기때문에이름에gydF4y2Ba |
“expected-improvement-per-second-plus”gydF4y2Ba |
MaxObjectiveEvaluationsgydF4y2Ba |
목적함수실행의최대횟수입니다。gydF4y2Ba | “bayesopt”gydF4y2Ba 또는gydF4y2Ba“randomsearch”gydF4y2Ba 의경우gydF4y2Ba30.gydF4y2Ba 이고,gydF4y2Ba“gridsearch”gydF4y2Ba 의경우그리드전체입니다。gydF4y2Ba |
MaxTimegydF4y2Ba |
시간제한으로,양의실수로지정됩니다。시간제한은초단위이며,gydF4y2Ba |
正gydF4y2Ba |
NumGridDivisionsgydF4y2Ba |
“gridsearch”gydF4y2Ba 의경우,각차원의값개수입니다。이값은각차원에대한값의개수를제공하는양의정수로구성된벡터또는모든차원에적용되는스칼라일수있습니다。이필드는범주형변수의경우무시됩니다。gydF4y2Ba |
10gydF4y2Ba |
ShowPlotsgydF4y2Ba |
플롯표시여부를나타내는논리값입니다。gydF4y2Ba真正的gydF4y2Ba 인경우,이필드는반복횟수에대해가장적합한목적함수값을플로팅합니다。하나또는두개의최적화모수가있고gydF4y2Ba优化器gydF4y2Ba 가gydF4y2Ba“bayesopt”gydF4y2Ba 인경우,gydF4y2BaShowPlotsgydF4y2Ba 는이모수에대해서도목적함수의모델을플로팅합니다。gydF4y2Ba |
真正的gydF4y2Ba |
SaveIntermediateResultsgydF4y2Ba |
优化器gydF4y2Ba 가gydF4y2Ba“bayesopt”gydF4y2Ba 인경우결과를저장할지여부를나타내는논리값입니다。gydF4y2Ba真正的gydF4y2Ba 인경우,이필드는각반복마다gydF4y2Ba“BayesoptResults”gydF4y2Ba 라는이름의작업공간변수를덮어씁니다。변수는gydF4y2BaBayesianOptimizationgydF4y2Ba 객체입니다。gydF4y2Ba |
假gydF4y2Ba |
详细的gydF4y2Ba |
명령줄에대한표시입니다。gydF4y2Ba
자세한내용은gydF4y2Ba |
1gydF4y2Ba |
UseParallelgydF4y2Ba |
베이즈최적화를병렬로실행할지여부를나타내는논리값으로,并行计算工具箱가필요합니다。병렬시간재현이불가능하기때문에,병렬베이즈최적화에서반드시재현가능한결과를산출하지는않습니다。자세한내용은gydF4y2Ba平行的贝叶斯优化gydF4y2Ba항목을참조하십시오。gydF4y2Ba | 假gydF4y2Ba |
重新分区gydF4y2Ba |
매반복시교차검증을다시분할할지여부를나타내는논리값입니다。gydF4y2Ba
|
假gydF4y2Ba |
다음과 같은 3.개 필드 이름 중 하나만 사용합니다.gydF4y2Ba | ||
CVPartitiongydF4y2Ba |
cvpartitiongydF4y2Ba 으로생성되는gydF4y2BacvpartitiongydF4y2Ba 객체입니다。gydF4y2Ba |
교차검증필드를지정하지않을경우gydF4y2Ba“Kfold”,5gydF4y2Ba |
坚持gydF4y2Ba |
홀드아웃비율을나타내는범위gydF4y2Ba(0,1)gydF4y2Ba 내스칼라입니다。gydF4y2Ba |
|
KfoldgydF4y2Ba |
1보다큰정수입니다。gydF4y2Ba |
예:gydF4y2Ba“HyperparameterOptimizationOptions”、结构(MaxObjectiveEvaluations, 60)gydF4y2Ba
데이터형:gydF4y2Ba结构体gydF4y2Ba
MdlgydF4y2Ba
— 훈련된 经济合作모델gydF4y2BaClassificationECOCgydF4y2Ba
모델객체|gydF4y2Ba紧凑分类gydF4y2Ba
모델객체|gydF4y2BaClassificationPartitionedECOCgydF4y2Ba
교차 검증된 모델 객체 |gydF4y2BaClassificationPartitionedLinearECOCgydF4y2Ba
교차 검증된 모델 객체 |gydF4y2Ba分类分区核函数gydF4y2Ba
교차 검증된 모델 객체gydF4y2Ba훈련된ECOC분류기로,모델객체gydF4y2BaClassificationECOCgydF4y2Ba
또는gydF4y2Ba紧凑分类gydF4y2Ba
나교차검증된모델객체gydF4y2BaClassificationPartitionedECOCgydF4y2Ba
,gydF4y2BaClassificationPartitionedLinearECOCgydF4y2Ba
또는gydF4y2Ba分类分区核函数gydF4y2Ba
로반환됩니다。gydF4y2Ba
다음표에서는gydF4y2BafitcecocgydF4y2Ba
에서 반환된 모델 객체의 유형이 사용자가 지정하는 이진 학습기 유형과 교차 검증 수행 여부에 따라 어떻게 달라지는지를 보여줍니다.gydF4y2Ba
선형분류모델학습기gydF4y2Ba | 커널분류모델학습기gydF4y2Ba | 교차검증gydF4y2Ba | 반환되는모델객체gydF4y2Ba |
---|---|---|---|
아니요gydF4y2Ba | 아니요gydF4y2Ba | 아니요gydF4y2Ba | ClassificationECOCgydF4y2Ba |
아니요gydF4y2Ba | 아니요gydF4y2Ba | 예gydF4y2Ba | ClassificationPartitionedECOCgydF4y2Ba |
예gydF4y2Ba | 아니요gydF4y2Ba | 아니요gydF4y2Ba | 紧凑分类gydF4y2Ba |
예gydF4y2Ba | 아니요gydF4y2Ba | 예gydF4y2Ba | ClassificationPartitionedLinearECOCgydF4y2Ba |
아니요gydF4y2Ba | 예gydF4y2Ba | 아니요gydF4y2Ba | 紧凑分类gydF4y2Ba |
아니요gydF4y2Ba | 예gydF4y2Ba | 예gydF4y2Ba | 分类分区核函数gydF4y2Ba |
HyperparameterOptimizationResultsgydF4y2Ba
- - - - - -하이퍼파라미터에대한교차검증최적화와관련된설명gydF4y2BaBayesianOptimizationgydF4y2Ba
객체|하이퍼파라미터및관련값으로구성된테이블gydF4y2Ba하이퍼파라미터에대한교차검증최적화와관련된설명으로,gydF4y2BaBayesianOptimizationgydF4y2Ba
객체또는하이퍼파라미터및관련값으로구성된테이블로반환됩니다。gydF4y2BaHyperparameterOptimizationResultsgydF4y2Ba
이름——값쌍의인수는gydF4y2BaOptimizeHyperparametersgydF4y2Ba
이름-값 쌍의 인수가 비어 있지 않은 경우에 비어 있지 않으며gydF4y2Ba学习者gydF4y2Ba
이름——값쌍의인수는선형이진학습기또는커널이진학습기를지정합니다。값은gydF4y2Ba超参数优化选项gydF4y2Ba
이름——값쌍의인수의설정에따라결정됩니다。gydF4y2Ba
“bayesopt”gydF4y2Ba
(디폴트 값) —gydF4y2BaBayesianOptimizationgydF4y2Ba
클래스의객체gydF4y2Ba
“gridsearch”gydF4y2Ba
또는gydF4y2Ba“randomsearch”gydF4y2Ba
- - - - - -사용된하이퍼파라미터,관측된목적함수값(교차검증손실),그리고관측값순위가가장낮은값(최상)에서가장높은값(최하)순으로포함된테이블gydF4y2Ba
데이터형:gydF4y2Ba表格gydF4y2Ba
fitcecocgydF4y2Ba
함수는선형분류모델을훈련시키는데에만희소행렬을지원합니다。기타모든모델의경우,비희소행렬을예측변수데이터로제공하십시오。gydF4y2Ba
이진손실gydF4y2Ba은이진학습기가관측값을클래스로얼마나잘분류하는지를결정하는분류점수와클래스의함수입니다。gydF4y2Ba
다음과같이가정하겠습니다。gydF4y2Ba
米gydF4y2BakjgydF4y2Ba는코딩설계행렬M의요소(k, j)입니다(즉,이진학습기j의k클래스에대응되는코드)。gydF4y2Ba
年代gydF4y2BajgydF4y2Ba는관측값에대한이진학습j의기점수입니다。gydF4y2Ba
g는이진손실함수입니다。gydF4y2Ba
는관측값에대한예측클래스입니다。gydF4y2Ba
손실기반디코딩gydF4y2Ba(에스칼레라(Escalera)등]gydF4y2Ba에서는이진학습기에대한이진손실의최소합을생성하는클래스가관측값에대한예측클래스를결정합니다(즉,gydF4y2Ba
).gydF4y2Ba
손실가중디코딩gydF4y2Ba(에스칼레라(Escalera)등]gydF4y2Ba에서는이진학습기에대한이진손실의최소평균을생성하는클래스가관측값에대한예측클래스를결정합니다(즉,gydF4y2Ba
).gydF4y2Ba올웨인등(Allwein)gydF4y2Ba에따르면손실가중디코딩은모든클래스에대한손실값을동일한동적범위로유지하여분류정확도를향상시킨다고합니다。gydF4y2Ba
다음표에는지원되는손실함수가요약되어있습니다。여기서ygydF4y2BajgydF4y2Ba는 특정 이진 학습기에 대한 클래스 레이블(이 세트에서는 {-1,1,0})이며, sgydF4y2BajgydF4y2Ba는 관측값 J에 대한 점수이며, g(y)gydF4y2BajgydF4y2Ba,年代gydF4y2BajgydF4y2Ba)입니다。gydF4y2Ba
값gydF4y2Ba | 설명gydF4y2Ba | 점수범위gydF4y2Ba | g (y)gydF4y2BajgydF4y2Ba,年代gydF4y2BajgydF4y2Ba)gydF4y2Ba |
---|---|---|---|
“binodeviance”gydF4y2Ba |
이항이탈도gydF4y2Ba | (-∞∞)gydF4y2Ba | 日志[1+exp(-2年)gydF4y2BajgydF4y2Ba年代gydF4y2BajgydF4y2Ba日志(2)])]/ [2gydF4y2Ba |
“指数”gydF4y2Ba |
지수(指数)gydF4y2Ba | (-∞∞)gydF4y2Ba | exp(可能是gydF4y2BajgydF4y2Ba年代gydF4y2BajgydF4y2Ba) / 2gydF4y2Ba |
“汉明”gydF4y2Ba |
해밍gydF4y2Ba | [0, 1]또는(-∞∞)gydF4y2Ba | [1–符号(y)gydF4y2BajgydF4y2Ba年代gydF4y2BajgydF4y2Ba) / 2gydF4y2Ba |
“枢纽”gydF4y2Ba |
경첩gydF4y2Ba | (-∞∞)gydF4y2Ba | 最大值(0,1–ygydF4y2BajgydF4y2Ba年代gydF4y2BajgydF4y2Ba) / 2gydF4y2Ba |
“线性”gydF4y2Ba |
선형gydF4y2Ba | (-∞∞)gydF4y2Ba | (1 - ygydF4y2BajgydF4y2Ba年代gydF4y2BajgydF4y2Ba) / 2gydF4y2Ba |
分对数的gydF4y2Ba |
로지스틱gydF4y2Ba | (-∞∞)gydF4y2Ba | 日志(1 + exp (- ygydF4y2BajgydF4y2Ba年代gydF4y2BajgydF4y2Ba日志(2)])]/ [2gydF4y2Ba |
“二次”gydF4y2Ba |
2차gydF4y2Ba | [0, 1]gydF4y2Ba | [1 - ygydF4y2BajgydF4y2Ba(2 sgydF4y2BajgydF4y2Ba- 1))gydF4y2Ba2gydF4y2Ba/2gydF4y2Ba |
소프트웨어는ygydF4y2BajgydF4y2Ba0.5 = 0일때손실이가되도록이진손실을정규화하고이진학습기의평균을사용하여총합을구합니다gydF4y2Ba[올웨인(Allwein)등]gydF4y2Ba.gydF4y2Ba
이진 손실을 전체 분류 손실(gydF4y2Ba损失gydF4y2Ba
객체함수와gydF4y2Ba预测gydF4y2Ba
객체함수의gydF4y2Ba“LossFun”gydF4y2Ba
이름——값쌍의인수로지정됨)로혼동하지마십시오。전체분류손실은ECOC분류기가전체적으로얼마나성능이좋은지를측정합니다。gydF4y2Ba
코딩설계gydF4y2Ba는어떤클래스가각각의이진학습기로훈련되는지,즉다중클래스문제가일련의이진문제로어떤식으로간소화되는지를요소가나타내는행렬입니다。gydF4y2Ba
코딩설계의각행은고유한클래스에대응되며,각열은이진학습기에대응됩니다。삼진코딩설계에서는특정열(또는이진학습기)에대해다음사항이적용됩니다。gydF4y2Ba
1을포함하는행은이진학습기에게대응클래스에포함된모든관측값을양성클래스로그룹화하라고지시합니다。gydF4y2Ba
1을포함하는행은이진학습기에게대응클래스에포함된모든관측값을음성클래스로그룹화하라고지시합니다。gydF4y2Ba
0을포함하는행은이진학습기에게대응클래스에포함된모든관측값을무시하라고지시합니다。gydF4y2Ba
해밍측정법에기반한최소쌍별행거리가큰값으로구성된코딩설계행렬이적합합니다。쌍별행거리에대한자세한내용은gydF4y2Ba확률코딩설계행렬gydF4y2Ba항목과gydF4y2Ba[4]gydF4y2Ba항목을참조하십시오。gydF4y2Ba
다음표에는많이사용되는코딩설계가설명되어있습니다。gydF4y2Ba
코딩설계gydF4y2Ba | 설명gydF4y2Ba | 학습기수gydF4y2Ba | 최소 쌍별 행 거리gydF4y2Ba |
---|---|---|---|
일대전부(卵子)gydF4y2Ba | 각각의이진학습기에서한클래스는양성이고나머지는음성입니다。이설계는모든양성클래스할당조합을사용합니다。gydF4y2Ba | KgydF4y2Ba | 2gydF4y2Ba |
일대일(蛋)gydF4y2Ba | 각각의이진학습기에서한클래스는양성이고,다른한클래스는음성이며,나머지클래스는무시됩니다。이설계는모든클래스쌍할당조합을사용합니다。gydF4y2Ba | K (K - 1) / 2gydF4y2Ba |
1gydF4y2Ba |
완전이진gydF4y2Ba | 이설계는클래스를모든이진조합으로분할하며어떠한클래스도무시하지않습니다。즉,모든클래스는gydF4y2Ba |
2gydF4y2BaK - 1gydF4y2Ba– 1gydF4y2Ba | 2gydF4y2BaK - 2gydF4y2Ba |
완전삼진gydF4y2Ba | 이설계는클래스를모든삼진(三元)조합으로분할합니다。즉,모든클래스는gydF4y2Ba |
(3gydF4y2BaKgydF4y2Ba– 2gydF4y2BaK + 1gydF4y2Ba+ 1) / 2gydF4y2Ba |
3.gydF4y2BaK - 2gydF4y2Ba |
순서형gydF4y2Ba | 첫번째이진학습기에서는첫번째클래스가음성이고나머지는양성입니다。두번째이진학습기에서는처음두개의클래스가음성이고나머지는양성이되는식입니다。gydF4y2Ba | K - 1gydF4y2Ba | 1gydF4y2Ba |
조밀확률gydF4y2Ba | 각각의 이진 학습기에서 소프트웨어는 클래스를 양성 클래스 또는 음성 클래스로 임의로 할당하되 적어도 하나의 양성 및 음성 클래스가 있도록 합니다. 자세한 내용은gydF4y2Ba확률코딩설계행렬gydF4y2Ba항목을참조하십시오。gydF4y2Ba | 임의(하지만 대략적으로 10原木gydF4y2Ba2gydF4y2BaK)gydF4y2Ba |
가변gydF4y2Ba |
희소확률gydF4y2Ba | 각각의이진학습기에서소프트웨어는클래스를각각0.25의확률로양성또는음성으로임의로할당하고0.5의확률로무시합니다。자세한내용은gydF4y2Ba확률코딩설계행렬gydF4y2Ba항목을참조하십시오。gydF4y2Ba | 임의(하지만 대략적으로 15原木gydF4y2Ba2gydF4y2BaK)gydF4y2Ba |
가변gydF4y2Ba |
이 플롯은 K가 증가함에 따라 코딩 설계에 필요한 이진 학습기 개수를 비교합니다.gydF4y2Ba
오류수정출력코드(ECOC)모델gydF4y2Ba은세개이상의클래스를갖는분류문제를이진분류기문제세트로간소화합니다。gydF4y2Ba
ECOC분류에는이진학습기가훈련하는데기준이되는클래스를결정하는코딩설계와이진분류기의결과(예측값)가어떻게집계되는지를결정하는디코딩체계가필요합니다。gydF4y2Ba
다음과같이가정하겠습니다。gydF4y2Ba
분류문제에는3개의클래스가있습니다。gydF4y2Ba
코딩설계가일대일입니다。3개의클래스에대해이코딩설계는다음과같습니다。gydF4y2Ba
디코딩체계가손실함g수를사용합니다。gydF4y2Ba
학습기가SVM입니다。gydF4y2Ba
ECOC알고리즘은다음단계를따라이분류모델을생성합니다。gydF4y2Ba
학습기1은클래스1또는클래스2에있는관측값에대해훈련하고,클래스1을양성클래스로,클래스2를음성클래스로처리합니다。다른학습기도비슷하게훈련됩니다。gydF4y2Ba
M을요소MgydF4y2Ba吉隆坡gydF4y2Ba을 포함하는 코딩 설계 행렬이라 하고, sgydF4y2BalgydF4y2Ba은학습기l의양성클래스에대한예측분류점수라고하겠습니다。알고리즘은새관측값을L이진학습기의손실에대한집계를최소화하는클래스(gydF4y2Ba )에할당합니다。gydF4y2Ba
ECOC모델은기타다중클래스모델에비해분류정확도를향상시킬수있습니다gydF4y2Ba[2]gydF4y2Ba.gydF4y2Ba
이진 학습기 개수는 클래스 개수에 따라 늘어납니다. 클래스가 많은 문제의 경우,gydF4y2BabinarycompletegydF4y2Ba
및gydF4y2BaternarycompletegydF4y2Ba
코딩설계는효율적이지않습니다。그러나다음을참고하십시오。gydF4y2Ba
K≤4이면gydF4y2BasparserandomgydF4y2Ba
대신gydF4y2BaternarycompletegydF4y2Ba
코딩설계를사용하십시오。gydF4y2Ba
K≤5이면gydF4y2BadenserandomgydF4y2Ba
대신gydF4y2BabinarycompletegydF4y2Ba
코딩설계를사용하십시오。gydF4y2Ba
명령창에gydF4y2BaMdl。CodingMatrix
를입력하여훈련된ECOC분류기의코딩설계행렬을표시할수있습니다。gydF4y2Ba
사용자는코딩행렬생성시응용방법에대한깊은지식을이용하고계산적인제약조건을고려해야합니다。충분한계산성능과시간이있다면여러코딩행렬을사용해본후가장성능이좋은코딩행렬을선택하십시오(즉,gydF4y2BaconfusionchartgydF4y2Ba
를사용하여각모델에대해정오분류표를확인하십시오)。gydF4y2Ba
리브-원아웃교차검증(gydF4y2BaLeaveoutgydF4y2Ba
)은관측값이많은데이터세트에대해서는효율적이지않습니다。대신k겹교차검증(gydF4y2BaKFoldgydF4y2Ba
)을사용하십시오。gydF4y2Ba
모델을훈련시킨후에는새데이터에대한레이블을예측하는C / c++코드를생성할수있습니다。C / c++코드를생성하려면gydF4y2BaMATLAB编码器™gydF4y2Ba가 필요합니다. 자세한 내용은gydF4y2Ba代码生成简介gydF4y2Ba항목을참조하십시오。gydF4y2Ba
사용자지정코딩행렬은특정형식을가져야합니다。소프트웨어는다음의조건을확보해사용자지정코딩행렬을검증합니다。gydF4y2Ba
모든요소는1,0또는1입니다。gydF4y2Ba
모든열이과1을각각하나이상포함합니다。gydF4y2Ba
모든고유한열벡터u및v에대해u≠v이고≠u - v입니다。gydF4y2Ba
모든행벡터가고유합니다。gydF4y2Ba
행렬이두클래스를분리할수있습니다。즉,다음규칙을따라행간을이동할수있습니다。gydF4y2Ba
1 1 1에서로또는에서1로수직으로이동할수있습니다。gydF4y2Ba
0이아닌요소간을수평으로이동할수있습니다。gydF4y2Ba
세로이동시행렬의열은단한번만사용할수있습니다。gydF4y2Ba
이러한규칙을사용하여행我에서행j로이동할수없다면我클래스와j클래스는해당설계행렬에의해분리될수없다는걸의미합니다。예를들어,다음코딩설계에서gydF4y2Ba
클래스 1.및 클래스 2.는 클래스 3.및 클래스 4.와 분리될 수 없습니다(즉, 열 2.에 0이 있으므로 행 2.의 -1.에서 열 2.로 가로로 이동할 수 없음). 따라서 소프트웨어가 이 코딩 설계를 거부합니다.gydF4y2Ba
병렬연산을사용하면(gydF4y2Ba选项gydF4y2Ba
참조),gydF4y2BafitcecocgydF4y2Ba
가이진학습기를병렬로훈련시킵니다。gydF4y2Ba
사전확률-소프트웨어는각각의이진학습기에대해지정된클래스사전확률(gydF4y2Ba之前gydF4y2Ba
)을정규화합니다。M을코딩설계행렬이라고하고,我(A, c)를표시행렬이라고하겠습니다。표시행렬은一와차원크기가같습니다。c一个의대응요소가이면표시행렬은값이1인요소를가지며,그렇지않은경우는요소값이0이됩니다。米gydF4y2Ba+1gydF4y2Ba과米gydF4y2Ba-1gydF4y2Ba을다음을충족하는K×L행렬이라고하겠습니다。gydF4y2Ba
米gydF4y2Ba+1gydF4y2Ba= M○我(M, 1)。여기서○은요소별곱셈(즉,gydF4y2BaMplus = M *(M == 1)gydF4y2Ba
)을 나타냅니다. 또한,gydF4y2Ba
을米gydF4y2Ba+1gydF4y2Ba의열벡터l이라고하겠습니다。gydF4y2Ba
米gydF4y2Ba-1gydF4y2Ba我(M, 1) = - M○(즉,gydF4y2BaMminus = - m。* (M = = 1)gydF4y2Ba
).또한,gydF4y2Ba
을米gydF4y2Ba-1gydF4y2Ba의열벡터l이라고하겠습니다。gydF4y2Ba
및gydF4y2Ba
라고 하겠습니다. 여기서 π는 지정된 클래스 사전 확률(gydF4y2Ba之前gydF4y2Ba
)의벡터입니다。gydF4y2Ba
그러면,이진학습l기에대한양성및음성스칼라클래스사전확률은다음과같습니다。gydF4y2Ba
여기서 j={-1,1}및gydF4y2Ba 은의1 -노름입니다。gydF4y2Ba
비용-소프트웨어는각이진학습기에대해K×K비용행렬C (gydF4y2Ba成本gydF4y2Ba
)를 정규화합니다. 이진 학습기 L에 대해 음성 클래스 관측값을 양성 클래스로 분류하는 비용은 다음과 같습니다.gydF4y2Ba
마찬가지로,양성클래스관측값을음성클래스로분류하는비용은다음과같습니다。gydF4y2Ba
l이진학습기에대한비용행렬은다음과같습니다。gydF4y2Ba
ECOC모델은오분류비용을클래스사전확률에통합하여수용합니다。gydF4y2Ba之前gydF4y2Ba
와gydF4y2Ba成本gydF4y2Ba
를지정하면소프트웨어가다음과같이클래스사전확률을조정합니다。gydF4y2Ba
클래스의개수K개가로주어지면다음과같이확률코딩설계행렬을생성합니다。gydF4y2Ba
다음행렬중하나를생성합니다。gydF4y2Ba
조밀확률- K×LgydF4y2BadgydF4y2Ba코딩설계행렬의각요소에동일한확률로1또는1을할당합니다。여기서gydF4y2Ba 입니다。gydF4y2Ba
희소 확률 — K×LgydF4y2Ba年代gydF4y2Ba코딩설계행렬의각요소에확률0.25로1또는1을할당하고확률0.5로0을할당합니다。여기서gydF4y2Ba 입니다。gydF4y2Ba
열이과1을각각하나이상포함하지않을경우소프트웨어는해당열을제거합니다。gydF4y2Ba
와서로다른열u v에대해u = v또는u = v이면소프트웨어는코딩설계행렬에서v를제거합니다。gydF4y2Ba
기본적으로10000개의행렬을임의로생성한다음아래의해밍측정법(gydF4y2Ba[4]gydF4y2Ba)에기반하여가장큰최소쌍별행거리를가지는행렬을유지합니다。gydF4y2Ba
여기서米gydF4y2BakgydF4y2BajgydF4y2BalgydF4y2Ba은코딩설계행렬j의요소입니다。gydF4y2Ba
기본적으로 효율성을 높이기 위해gydF4y2BafitcecocgydF4y2Ba
는모든선형SVM이진학습기에대해속성gydF4y2BaαgydF4y2Ba
,gydF4y2Ba万博1manbetxSupportVectorLabelsgydF4y2Ba
,gydF4y2Ba万博1manbetxSupportVectorsgydF4y2Ba
를비워둡니다。gydF4y2BafitcecocgydF4y2Ba
는 모델 표시 화면에gydF4y2BaαgydF4y2Ba
가아니라gydF4y2BaβgydF4y2Ba
를표시합니다。gydF4y2Ba
αgydF4y2Ba
,gydF4y2Ba万博1manbetxSupportVectorLabelsgydF4y2Ba
,gydF4y2Ba万博1manbetxSupportVectorsgydF4y2Ba
를저장하려면gydF4y2BafitcecocgydF4y2Ba
에서포트벡터를저장함을지정하는선형SVM템플릿을전달하십시오。예를들어,다음을입력합니다。gydF4y2Ba
t=模板SVM(gydF4y2Ba“Save万博1manbetxSupportVectors”gydF4y2BaMdl = fitcecoc(X,Y,gydF4y2Ba“学习者”gydF4y2Bat);gydF4y2Ba
결과로생성된gydF4y2BaClassificationECOCgydF4y2Ba
모델을gydF4y2Badiscard万博1manbetxSupportVectorsgydF4y2Ba
로전달하여서포트벡터및관련값을제거할수있습니다。gydF4y2Ba
[1] Allwein,E.,R.Schapire和Y.Singer.“将多类简化为二类:边际分类的统一方法”,《机器学习研究杂志》,2000年第1卷,第113-141页。gydF4y2Ba
[2] Fürnkranz,约翰内斯,《轮循分类》。j·马赫。学习。参考文献,第2卷,2002年,第721-747页。gydF4y2Ba
Pujol, S. Escalera, S. O. Pujol, P. Radeva。《论三元纠错输出码的译码过程》。模式分析与机器智能学报。2010年第32卷第7期120-134页。gydF4y2Ba
[4] 《用于纠错输出码稀疏设计的三元码的可分性》,《模式记录》,第30卷,2009年第3期,第285-297页。gydF4y2Ba
사용법관련참고및제한사항:gydF4y2Ba
지원되는 구문은 다음과 같습니다.gydF4y2Ba
Mdl = fitcecoc (X, Y)gydF4y2Ba
Mdl = fitcecoc (X, Y,名称,值)gydF4y2Ba
(Mdl FitInfo HyperparameterOptimizationResults] = fitcecoc (X, Y,名称,值)gydF4y2Ba
- - - - - -gydF4y2BafitcecocgydF4y2Ba
는gydF4y2Ba“OptimizeHyperparameters”gydF4y2Ba
이름——값쌍의인수를지정할경우출력인수gydF4y2BaFitInfogydF4y2Ba
와gydF4y2BaHyperparameterOptimizationResultsgydF4y2Ba
도추가로반환합니다。gydF4y2Ba
FitInfogydF4y2Ba
출력인수는향후사용을위해현재예약되어있는빈구조체형배열입니다。gydF4y2Ba
교차검증과관련된옵션은지원되지않습니다。지원되는이름——값쌍의인수는다음과같습니다。gydF4y2Ba
“类名”gydF4y2Ba
“成本”gydF4y2Ba
“编码”gydF4y2Ba
— 디폴트 값은gydF4y2Ba“onevsall”gydF4y2Ba
입니다。gydF4y2Ba
“HyperparameterOptimizationOptions”gydF4y2Ba
— 교차 검증의 경우 高的형에 대한 최적화에는gydF4y2Ba“坚持”gydF4y2Ba
검증만지원됩니다。기본적으로소프트웨어는데이터의20%를선택하여홀드아웃검증데이터로남겨두고나머지데이터를사용하여모델을훈련시킵니다。이인수를사용하여홀드아웃비율에다른값을지정할수있습니다。예를들어,데이터의30%를검증데이터로남겨두려면gydF4y2Ba“HyperparameterOptimizationOptions”、结构(“抵抗”,0.3)gydF4y2Ba
을 지정하십시오.gydF4y2Ba
“学习者”gydF4y2Ba
— 디폴트 값은gydF4y2Ba“线性”gydF4y2Ba
입니다。gydF4y2Ba“线性”gydF4y2Ba
,gydF4y2Ba“内核”gydF4y2Ba
,gydF4y2BatemplateLineargydF4y2Ba
또는gydF4y2Ba模板核gydF4y2Ba
객체나 이러한 객체로 구성된 셀형 배열을 지정할 수 있습니다.gydF4y2Ba
“OptimizeHyperparameters”gydF4y2Ba
— 선형 이진 학습기를 사용할 경우gydF4y2Ba“正规化”gydF4y2Ba
하이퍼파라미터의 값은gydF4y2Ba“岭”gydF4y2Ba
여야합니다。gydF4y2Ba
“之前”gydF4y2Ba
“详细”gydF4y2Ba
— 디폴트 값은gydF4y2Ba1gydF4y2Ba
입니다。gydF4y2Ba
“重量”gydF4y2Ba
다음의추가적인이름——값쌍의인수는高형배열에만사용할수있습니다。gydF4y2Ba
“NumConcurrent”gydF4y2Ba
——파일I / O작업을결합하여동시에훈련시키는이진학습기개수를지정하는양의정수스칼라입니다。gydF4y2Ba“NumConcurrent”gydF4y2Ba
의디폴트값은gydF4y2Ba1gydF4y2Ba
이며,이경우gydF4y2BafitcecocgydF4y2Ba
는이진분류기를순차적으로훈련시킵니다。입력배열을분산군집메모리에담을수없는경우gydF4y2Ba“NumConcurrent”gydF4y2Ba
가 가장 유용합니다. 그렇지 않으면 입력 배열이 캐시될 수도 있으며 속도 향상은 거의 이뤄지지 않습니다.gydF4y2Ba
Apache火花™에서코드를실행하면통신에사용가능한메모리에따라gydF4y2BaNumConcurrentgydF4y2Ba
의상한값이제한됩니다。Apache火花구성에서gydF4y2Ba“spark.executor.memory”gydF4y2Ba
속성과gydF4y2Ba“spark.driver.memory”gydF4y2Ba
속성을 확인하십시오. 자세한 내용은gydF4y2Baparallel.cluster.HadoopgydF4y2Ba
(并行计算工具箱)gydF4y2Ba항목을참조하십시오。Apache火花및코드가실행되는위치를제어하는기타실행환경에대한자세한내용은gydF4y2Ba扩展高数组与其他产品s manbetx 845gydF4y2Ba항목을참조하십시오。gydF4y2Ba
자세한내용은gydF4y2Ba高형배열gydF4y2Ba항목을참조하십시오。gydF4y2Ba
병렬로실행하려면다음방법중하나로gydF4y2Ba“UseParallel”gydF4y2Ba
옵션을gydF4y2Ba真正的gydF4y2Ba
로설정해야합니다。gydF4y2Ba
斯塔塞特gydF4y2Ba
을사용하여选项구조체의gydF4y2Ba“UseParallel”gydF4y2Ba
필드를gydF4y2Ba真正的gydF4y2Ba
로설정하고gydF4y2BafitceocgydF4y2Ba
에대한호출에gydF4y2Ba“选项”gydF4y2Ba
이름——값쌍의인수를지정합니다。gydF4y2Ba
예:gydF4y2Ba“选项”,statset('UseParallel',true)gydF4y2Ba
자세한내용은gydF4y2Ba“选项”gydF4y2Ba
이름——값쌍의인수를참조하십시오。gydF4y2Ba
병렬하이퍼파라미터최적화를수행하려면gydF4y2BafitceocgydF4y2Ba
에대한호출에gydF4y2Ba“HyperparameterOptions”、结构(UseParallel,真的)gydF4y2Ba
이름——값쌍의인수를사용하십시오。gydF4y2Ba
병렬하이퍼파라미터최적화에대한자세한내용은gydF4y2Ba平行的贝叶斯优化gydF4y2Ba항목을참조하십시오。gydF4y2Ba
ClassificationECOCgydF4y2Ba
|gydF4y2Ba紧凑分类gydF4y2Ba
|gydF4y2BaClassificationPartitionedECOCgydF4y2Ba
|gydF4y2Ba损失gydF4y2Ba
|gydF4y2Ba预测gydF4y2Ba
|gydF4y2BadesignecocgydF4y2Ba
|gydF4y2Ba斯塔塞特gydF4y2Ba
|gydF4y2BaClassificationPartitionedLinearECOCgydF4y2Ba
|gydF4y2Ba分类分区核函数gydF4y2Ba
이 예제의 수정된 버전이 있습니다. 사용자가 편집한 내용을 반영하여 이 예제를 여시겠습니까?gydF4y2Ba
다음 MATLAB명령에 해당하는 링크를 클릭했습니다.gydF4y2Ba
명령을실행하려면MATLAB명령창에입력하십시오。웹브라우저는MATLAB명령을지원하지않습니다。gydF4y2Ba
选择一个网站,在那里获得翻译的内容,并看到当地的活动和优惠。根据您的位置,我们建议您选择:gydF4y2Ba.gydF4y2Ba
选择gydF4y2Ba网站gydF4y2Ba您还可以从以下列表中选择网站:gydF4y2Ba
选择中国网站(中文或英文)以获得最佳网站性能。其他MathWorks国家站点没有针对您所在位置的访问进行优化。gydF4y2Ba