本文主要是介绍用Matlab 2015a svmtrain函数训练的SVM model在2021b无法使用的解决方法，希望对大家解决编程问题提供一定的参考价值，需要的开发者们随着小编来一起学习吧！
背景
- 与r2015a版本的Matlab相比，r2021b版本中包含更多集成好的算法模块（尤其是深度学习的模块），想把原来r2015a版本的代码升级到r2021b
- 高版本的Matlab已经采用fitcsvm函数和predict函数替代了旧版本中svmtrain函数和svmclassify函数。在r2021b中运行原来的代码时提示
未定义与 'struct' 类型的输入参数相对应的函数 'svmclassify'
- 当直接把svmclassify换成predict函数时，提示
错误使用 predict (第 124 行)
No valid system or dataset was specified.
- 原先用于训练svm model的数据已经丢失，无法用新版本的fitcsvm函数重新训练svm model，想直接在r2021b中调用原先训练好的svm model
解决方法
把下面这4个函数保存到原来的代码文件夹中，再在Matlab r2021b中运行原来的代码即可，注意运行原来的代码前，把 svmclassify 改成 svmclassify_r2015a。（这4个函数是从matlab r2015a中复制过来并修改了函数名的）
function outclass = svmclassify_r2015a(svmStruct,sample, varargin)
%SVMCLASSIFY Classify data using a support vector machine
% SVMCLASSIFY will be removed in a future release. Use the PREDICT method of
% an object returned by FITCSVM instead.
%
% GROUP = SVMCLASSIFY(SVMSTRUCT, TEST) classifies each row in TEST using
% the support vector machine classifier structure SVMSTRUCT created
% using SVMTRAIN, and returns the predicted class level GROUP. TEST must
% have the same number of columns as the data used to train the
% classifier in SVMTRAIN. GROUP indicates the group to which each row of
% TEST is assigned.
%
% GROUP = SVMCLASSIFY(...,'SHOWPLOT',true) plots the test data TEST on
% the figure created using the SHOWPLOT option in SVMTRAIN.
%
% Example:
% % Load the data and select features for classification
% load fisheriris
% X = [meas(:,1), meas(:,2)];
% % Extract the Setosa class
% Y = nominal(ismember(species,'setosa'));
% % Randomly partitions observations into a training set and a test
% % set using stratified holdout
% P = cvpartition(Y,'Holdout',0.20);
% % Use a linear support vector machine classifier
% svmStruct = svmtrain(X(P.training,:),Y(P.training),'showplot',true);
% C = svmclassify(svmStruct,X(P.test,:),'showplot',true);
% err_rate = sum(Y(P.test)~= C)/P.TestSize % mis-classification rate
% conMat = confusionmat(Y(P.test),C) % the confusion matrix
%
% See also SVMTRAIN, CLASSIFY, TREEBAGGER, fitcsvm.% Copyright 2004-2014 The MathWorks, Inc.% References:
%
% [1] Cristianini, N., Shawe-Taylor, J An Introduction to Support
% Vector Machines, Cambridge University Press, Cambridge, UK. 2000.
% http://www.support-vector.net
% [2] Kecman, V, Learning and Soft Computing,
% MIT Press, Cambridge, MA. 2001.
% [3] Suykens, J.A.K., Van Gestel, T., De Brabanter, J., De Moor, B.,
% Vandewalle, J., Least Squares Support Vector Machines,
% World Scientific, Singapore, 2002.% set defaults
plotflag = false;% check inputs
narginchk(2, Inf);% deal with struct input case
if ~isstruct(svmStruct)error(message('stats:svmclassify:TwoInputsNoStruct'));
endif ~isnumeric(sample) || ~ismatrix(sample)error(message('stats:svmclassify:BadSample'));
endif size(sample,2)~=size(svmStruct.SupportVectors,2)error(message('stats:svmclassify:TestSizeMismatch'));
end% deal with the various inputs
if nargin > 2if rem(nargin,2) == 1error(message('stats:svmclassify:IncorrectNumberOfArguments'));endokargs = {'showplot','-compilerhelper'};for j=1:2:nargin-2pname = varargin{j};pval = varargin{j+1};k = find(strncmpi(pname, okargs,numel(pname)));if isempty(k)error(message('stats:svmclassify:UnknownParameterName', pname));elseif length(k)>1error(message('stats:svmclassify:AmbiguousParameterName', pname));elseswitch(k)case 1 % plotflag ('SHOWPLOT')plotflag = opttf(pval,okargs{k}); case 2 % help the compiler find required function handles by including svmtrainsvmtrain_r2015a(eye(2),[1 0]);endendend
endgroupnames = svmStruct.GroupNames;% check group is a vector -- though char input is special...
if ~isvector(groupnames) && ~ischar(groupnames)error(message('stats:svmclassify:GroupNotVector'));
end% grp2idx sorts a numeric grouping var ascending, and a string grouping
% var by order of first occurrence
[~,groupString,glevels] = grp2idx(groupnames); % do the classification
if ~isempty(sample)% shift and scale the data if necessary:sampleOrig = sample;if ~isempty(svmStruct.ScaleData)for c = 1:size(sample, 2)sample(:,c) = svmStruct.ScaleData.scaleFactor(c) * ...(sample(:,c) + svmStruct.ScaleData.shift(c));endendtryoutclass = svmdecision_r2015a(sample,svmStruct);catch MEerror(message('stats:svmclassify:ClassifyFailed', ME.message));endif plotflagif isempty(svmStruct.FigureHandles)warning(message('stats:svmclassify:NoTrainingFigure'));elsetryhAxis = svmStruct.FigureHandles{1};hLines = svmStruct.FigureHandles{2};hSV = svmStruct.FigureHandles{3};% unscale the data for plotting purposes[~,hClassLines] = svmplotdata(sampleOrig,outclass,hAxis); trainingString = strcat(cellstr(groupString),' (training)');sampleString = strcat(cellstr(groupString),' (classified)');legendHandles = {hLines(1),hClassLines{1},...hLines(2),hClassLines{2},hSV};legendNames = {trainingString{1},sampleString{1},...trainingString{2},sampleString{2},'Support Vectors'};ok = ~cellfun(@isempty,legendHandles);legend([legendHandles{ok}],legendNames(ok));catch MEwarning(message('stats:svmclassify:DisplayFailed', ME.message));endendendoutclass(outclass == -1) = 2;unClassified = isnan(outclass);outclass = glevels(outclass(~unClassified),:);if any(unClassified)tryoutclass = statinsertnan(unClassified,outclass);catch MEif ~isequal(ME.identifier,'stats:statinsertnan:LogicalInput')rethrow(ME);elseerror(message('stats:svmclassify:logicalwithNaN'));endendendelseoutclass = [];
endfunction [svm_struct, svIndex] = svmtrain_r2015a(training, groupnames, varargin)
%SVMTRAIN Train a support vector machine classifier
% SVMTRAIN will be removed in a future release. Use FITCSVM instead.
%
% SVMSTRUCT = SVMTRAIN(TRAINING, Y) trains a support vector machine (SVM)
% classifier on data taken from two groups. TRAINING is a numeric matrix
% of predictor data. Rows of TRAINING correspond to observations; columns
% correspond to features. Y is a column vector that contains the known
% class labels for TRAINING. Y is a grouping variable, i.e., it can be a
% categorical, numeric, or logical vector; a cell vector of strings; or a
% character matrix with each row representing a class label (see help for
% groupingvariable). Each element of Y specifies the group the
% corresponding row of TRAINING belongs to. TRAINING and Y must have the
% same number of rows. SVMSTRUCT contains information about the trained
% classifier, including the support vectors, that is used by SVMCLASSIFY
% for classification. SVMTRAIN treats NaNs, empty strings or 'undefined'
% values as missing values and ignores the corresponding rows in
% TRAINING and Y.
%
% SVMSTRUCT = SVMTRAIN(TRAINING, Y, 'PARAM1',val1, 'PARAM2',val2, ...)
% specifies one or more of the following name/value pairs:
%
% Name Value
% 'kernel_function' A string or a function handle specifying the
% kernel function used to represent the dot
% product in a new space. The value can be one of
% the following:
% 'linear' - Linear kernel or dot product
% (default). In this case, SVMTRAIN
% finds the optimal separating plane
% in the original space.
% 'quadratic' - Quadratic kernel
% 'polynomial' - Polynomial kernel with default
% order 3. To specify another order,
% use the 'polyorder' argument.
% 'rbf' - Gaussian Radial Basis Function
% with default scaling factor 1. To
% specify another scaling factor,
% use the 'rbf_sigma' argument.
% 'mlp' - Multilayer Perceptron kernel (MLP)
% with default weight 1 and default
% bias -1. To specify another weight
% or bias, use the 'mlp_params'
% argument.
% function - A kernel function specified using
% @(for example @KFUN), or an
% anonymous function. A kernel
% function must be of the form
%
% function K = KFUN(U, V)
%
% The returned value, K, is a matrix
% of size M-by-N, where M and N are
% the number of rows in U and V
% respectively.
%
% 'rbf_sigma' A positive number specifying the scaling factor
% in the Gaussian radial basis function kernel.
% Default is 1.
%
% 'polyorder' A positive integer specifying the order of the
% polynomial kernel. Default is 3.
%
% 'mlp_params' A vector [P1 P2] specifying the parameters of MLP
% kernel. The MLP kernel takes the form:
% K = tanh(P1*U*V' + P2),
% where P1 > 0 and P2 < 0. Default is [1,-1].
%
% 'method' A string specifying the method used to find the
% separating hyperplane. Choices are:
% 'SMO' - Sequential Minimal Optimization (SMO)
% method (default). It implements the L1
% soft-margin SVM classifier.
% 'QP' - Quadratic programming (requires an
% Optimization Toolbox license). It
% implements the L2 soft-margin SVM
% classifier. Method 'QP' doesn't scale
% well for TRAINING with large number of
% observations.
% 'LS' - Least-squares method. It implements the
% L2 soft-margin SVM classifier.
%
% 'options' Options structure created using either STATSET or
% OPTIMSET.
% * When you set 'method' to 'SMO' (default),
% create the options structure using STATSET.
% Applicable options:
% 'Display' Level of display output. Choices
% are 'off' (the default), 'iter', and
% 'final'. Value 'iter' reports every
% 500 iterations.
% 'MaxIter' A positive integer specifying the
% maximum number of iterations allowed.
% Default is 15000 for method 'SMO'.
% * When you set method to 'QP', create the
% options structure using OPTIMSET. For details
% of applicable options choices, see QUADPROG
% options. SVM uses a convex quadratic program,
% so you can choose the 'interior-point-convex'
% algorithm in QUADPROG.
%
% 'tolkkt' A positive scalar that specifies the tolerance
% with which the Karush-Kuhn-Tucker (KKT) conditions
% are checked for method 'SMO'. Default is
% 1.0000e-003.
%
% 'kktviolationlevel' A scalar specifying the fraction of observations
% that are allowed to violate the KKT conditions for
% method 'SMO'. Setting this value to be positive
% helps the algorithm to converge faster if it is
% fluctuating near a good solution. Default is 0.
%
% 'kernelcachelimit' A positive scalar S specifying the size of the
% kernel matrix cache for method 'SMO'. The
% algorithm keeps a matrix with up to S * S
% double-precision numbers in memory. Default is
% 5000. When the number of points in TRAINING
% exceeds S, the SMO method slows down. It's
% recommended to set S as large as your system
% permits.
%
% 'boxconstraint' The box constraint C for the soft margin. C can be
% a positive numeric scalar or a vector of positive
% numbers with the number of elements equal to the
% number of rows in TRAINING.
% Default is 1.
% * If C is a scalar, it is automatically rescaled
% by N/(2*N1) for the observations of group one,
% and by N/(2*N2) for the observations of group
% two, where N1 is the number of observations in
% group one, N2 is the number of observations in
% group two. The rescaling is done to take into
% account unbalanced groups, i.e., when N1 and N2
% are different.
% * If C is a vector, then each element of C
% specifies the box constraint for the
% corresponding observation.
%
% 'autoscale' A logical value specifying whether or not to
% shift and scale the data points before training.
% When the value is true, the columns of TRAINING
% are shifted and scaled to have zero mean unit
% variance. Default is true.
%
% 'showplot' A logical value specifying whether or not to show
% a plot. When the value is true, SVMTRAIN creates a
% plot of the grouped data and the separating line
% for the classifier, when using data with 2
% features (columns). Default is false.
%
% SVMSTRUCT is a structure having the following properties:
%
% SupportVectors Matrix of data points with each row corresponding
% to a support vector.
% Note: when 'autoscale' is false, this field
% contains original support vectors in TRAINING.
% When 'autoscale' is true, this field contains
% shifted and scaled vectors from TRAINING.
% Alpha Vector of Lagrange multipliers for the support
% vectors. The sign is positive for support vectors
% belonging to the first group and negative for
% support vectors belonging to the second group.
% Bias Intercept of the hyperplane that separates
% the two groups.
% Note: when 'autoscale' is false, this field
% corresponds to the original data points in
% TRAINING. When 'autoscale' is true, this field
% corresponds to shifted and scaled data points.
% KernelFunction The function handle of kernel function used.
% KernelFunctionArgs Cell array containing the additional arguments
% for the kernel function.
% GroupNames A column vector that contains the known
% class labels for TRAINING. Y is a grouping
% variable (see help for groupingvariable).
% SupportVectorIndices A column vector indicating the indices of support
% vectors.
% ScaleData This field contains information about auto-scale.
% When 'autoscale' is false, it is empty. When
% 'autoscale' is set to true, it is a structure
% containing two fields:
% shift - A row vector containing the negative
% of the mean across all observations
% in TRAINING.
% scaleFactor - A row vector whose value is
% 1./STD(TRAINING).
% FigureHandles A vector of figure handles created by SVMTRAIN
% when 'showplot' argument is TRUE.
%
% Example:
% % Load the data and select features for classification
% load fisheriris
% X = [meas(:,1), meas(:,2)];
% % Extract the Setosa class
% Y = nominal(ismember(species,'setosa'));
% % Randomly partitions observations into a training set and a test
% % set using stratified holdout
% P = cvpartition(Y,'Holdout',0.20);
% % Use a linear support vector machine classifier
% svmStruct = svmtrain(X(P.training,:),Y(P.training),'showplot',true);
% C = svmclassify(svmStruct,X(P.test,:),'showplot',true);
% errRate = sum(Y(P.test)~= C)/P.TestSize %mis-classification rate
% conMat = confusionmat(Y(P.test),C) % the confusion matrix
%
% See also SVMCLASSIFY, CLASSIFY, TREEBAGGER, GROUPINGVARIABLE, fitcsvm.% Copyright 2004-2014 The MathWorks, Inc.% References:
%
% [1] Cristianini, N., Shawe-Taylor, J An Introduction to Support
% Vector Machines, Cambridge University Press, Cambridge, UK. 2000.
% http://www.support-vector.net
% [2] Kecman, V, Learning and Soft Computing,
% MIT Press, Cambridge, MA. 2001.
% [3] Suykens, J.A.K., Van Gestel, T., De Brabanter, J., De Moor, B.,
% Vandewalle, J., Least Squares Support Vector Machines,
% World Scientific, Singapore, 2002.
% [4] J.C. Platt: A Fast Algorithm for Training Support Vector
% Machines, Advances in Kernel Methods - Support Vector Learning,
% MIT Press, 1998.
% [5] J.C. Platt: Fast Training of Support Vector Machines using
% Sequential Minimal Optimization Microsoft Research Technical
% Report MSR-TR-98-14, 1998.
% [6] http://www.kernel-machines.org/papers/tr-30-1998.ps.gz
%
% SVMTRAIN(...,'KFUNARGS',ARGS) allows you to pass additional
% arguments to kernel functions.narginchk(2, Inf);% check group is a vector or a char array
if ~isvector(groupnames) && ~ischar(groupnames)error(message('stats:svmtrain:GroupNotVector'));
end
% make sure that the data are correctly oriented.
if size(groupnames,1) == 1groupnames = groupnames';
endif ~isnumeric(training) || ~ismatrix(training) error(message('stats:svmtrain:TrainingBadType'));
end% grp2idx sorts a numeric grouping var ascending, and a string grouping
% var by order of first occurrence
[groupIndex, groupString] = grp2idx(groupnames);% make sure data is the right size
if size(training,1) ~= size(groupIndex,1)if size(training,2) == size(groupIndex,1)training = training';elseerror(message('stats:svmtrain:DataGroupSizeMismatch'))end
endif isempty(training)error(message('stats:svmtrain:NoData'))
endnans = isnan(groupIndex) | any(isnan(training),2);
if any(nans)training(nans,:) = [];groupIndex(nans) = [];
end
if isempty(training)error(message('stats:svmtrain:NoData'))
endngroups = length(unique(groupIndex));
nPoints = length(groupIndex);if ngroups > 2error(message('stats:svmtrain:TooManyGroups', ngroups))
end
if length(groupString) > ngroupswarning(message('stats:svmtrain:EmptyGroups'));end
% convert to groupIndex from 2 to -1.
groupIndex = 1 - (2* (groupIndex-1));pnames = {'kernel_function','method','showplot', 'polyorder','mlp_params',...'boxconstraint','rbf_sigma','autoscale', 'options',...'tolkkt','kktviolationlevel','kernelcachelimit'...'kfunargs', 'quadprog_opts','smo_opts'};
dflts = { 'linear', [], false, [], [], ....1, [], true , [] , ....[], [], [],...{} , [] , []};
[kfun,optimMethod, plotflag, polyOrder, mlpParams, boxC, rbf_sigma, ...autoScale, opts, tolkkt, kktvl,kerCL, kfunargs, qpOptsInput, ...smoOptsInput] = internal.stats.parseArgs(pnames, dflts, varargin{:});usePoly = false;
useMLP = false;
useSigma = false;
%parse kernel functions
if ischar(kfun)okfuns = {'linear','quadratic', 'radial','rbf','polynomial','mlp'};[~,i] = internal.stats.getParamVal(kfun,okfuns,'kernel_function');switch icase 1kfun = @linear_kernel;case 2kfun = @quadratic_kernel;case {3,4}kfun = @rbf_kernel;useSigma = true;case 5kfun = @poly_kernel;usePoly = true;case 6kfun = @mlp_kernel;useMLP = true;end
elseif ~isa(kfun, 'function_handle')error(message('stats:svmtrain:BadKernelFunction'));
end%parse optimization method
optimList ={'QP','SMO','LS'};
i = 2; % set to 'SMO'if ~isempty(optimMethod)[~,i] = internal.stats.getParamVal(optimMethod,optimList,'Method');if i==1 && ( ~license('test', 'optimization_toolbox') ...|| isempty(which('quadprog')))warning(message('stats:svmtrain:NoOptim'));i = 2;end
endif i == 2 && ngroups==1error(message('stats:svmtrain:InvalidY'));
end
optimMethod = optimList{i};% The large scale solver cannot handle this type of problem, so turn it off.
% qp_opts = optimset('LargeScale','Off','display','off');
% We can use the 'interior-point-convex' option
qp_opts = optimset('Algorithm','interior-point-convex','display','off');
smo_opts = statset('Display','off','MaxIter',15000);
%parse opts. opts will override 'quadprog_opt' and 'smo_opt' argument
if ~isempty(opts)qp_opts = optimset(qp_opts,opts);smo_opts = statset(smo_opts,opts);
else% only consider undocumented 'quadprog_opts' arguments% when 'opts' is empty; Otherwise, ignore 'quadprog_opts'if ~isempty(qpOptsInput)if isstruct(qpOptsInput)qp_opts = optimset(qp_opts,qpOptsInput);elseif iscell(qpOptsInput)qp_opts = optimset(qp_opts,qpOptsInput{:});elseerror(message('stats:svmtrain:BadQuadprogOpts'));endend
end% Turn off deprecation warning for svmsmoset
warning('off','stats:obsolete:ReplaceThisWith');
cleanupObj = onCleanup(@() warning('on','stats:obsolete:ReplaceThisWith'));if ~isempty(smoOptsInput) && isempty(tolkkt) && isempty(kktvl) ...&& isempty(kerCL) && isempty(opts)%back-compatibility.smo_opts = svmsmoset(smoOptsInput);
elseif isempty(tolkkt)tolkkt = 1e-3;endif isempty(kerCL)kerCL = 5000;endif isempty(kktvl)kktvl = 0;endsmo_opts = svmsmoset(smo_opts,'tolkkt',tolkkt,'KernelCacheLimit',kerCL,....'KKTViolationLevel',kktvl);
endif ~isscalar(smo_opts.TolKKT) || ~isnumeric(smo_opts.TolKKT) || smo_opts.TolKKT <= 0error(message('stats:svmtrain:badTolKKT'));
endif ~isscalar(smo_opts.KKTViolationLevel) || ~isnumeric(smo_opts.KKTViolationLevel)...|| smo_opts.KKTViolationLevel < 0 || smo_opts.KKTViolationLevel > 1error(message('stats:svmtrain:badKKTVL'));
endif ~isscalar(smo_opts.KernelCacheLimit) || ~isnumeric(smo_opts.KernelCacheLimit)...||smo_opts.KernelCacheLimit < 0error(message('stats:svmtrain:badKerCL'));
end%parse plot flag
plotflag = opttf(plotflag,'showplot');
if plotflag && size(training,2) ~=2plotflag = false;warning(message('stats:svmtrain:OnlyPlot2D'));
endif ~isempty(kfunargs) && ~iscell(kfunargs)kfunargs = {kfunargs};
end%polyOrder
if ~isempty(polyOrder)%setPoly = true;if ~usePolywarning(message('stats:svmtrain:PolyOrderNotPolyKernel'));elsekfunargs = {polyOrder};end
end% mlpparams
if ~isempty(mlpParams)if ~isnumeric(mlpParams) || numel(mlpParams)~=2error(message('stats:svmtrain:BadMLPParams'));endif mlpParams(1) <= 0error(message('stats:svmtrain:MLPWeightNotPositive'))endif mlpParams(2) >= 0warning(message('stats:svmtrain:MLPBiasNotNegative'))endif ~useMLPwarning(message('stats:svmtrain:MLPParamNotMLPKernel'));elsekfunargs = {mlpParams(1), mlpParams(2)};end
end%rbf_sigma
if ~isempty(rbf_sigma)if useSigmakfunargs = {rbf_sigma};elsewarning(message('stats:svmtrain:RBFParamNotRBFKernel'))end
end% box constraint: it can be a positive numeric scalar or a numeric vector
% of the same length as the number of data points
if isscalar(boxC) && isnumeric(boxC) && boxC > 0% scalar input: adjust to group size and transform into vector% set default value of box constraintboxconstraint = ones(nPoints, 1); n1 = length(find(groupIndex==1));n2 = length(find(groupIndex==-1));c1 = 0.5 * boxC * nPoints / n1;c2 = 0.5 * boxC * nPoints / n2;boxconstraint(groupIndex==1) = c1;boxconstraint(groupIndex==-1) = c2;
elseif isvector(boxC) && isnumeric(boxC) && all(boxC > 0) && (length(boxC) == nPoints)% vector inputboxconstraint = boxC;
elseerror(message('stats:svmtrain:InvalidBoxConstraint'));
end
% If boxconstraint == Inf then convergence will not
% happen so fix the value to 1/sqrt(eps).
boxconstraint = min(boxconstraint,repmat(1/sqrt(eps(class(boxconstraint))),...size(boxconstraint)));autoScale = opttf(autoScale,'autoscale');% plot the data if requested
if plotflag[hAxis,hLines] = svmplotdata(training,groupIndex);hLines = [hLines{1} hLines{2}];legend(hLines,cellstr(groupString));
end% autoscale data if required,
scaleData = [];
if autoScalescaleData.shift = - mean(training);stdVals = std(training);scaleData.scaleFactor = 1./stdVals;% leave zero-variance data unscaled:scaleData.scaleFactor(~isfinite(scaleData.scaleFactor)) = 1;% shift and scale columns of data matrix:for c = 1:size(training, 2)training(:,c) = scaleData.scaleFactor(c) * ...(training(:,c) + scaleData.shift(c));end
endif strcmpi(optimMethod, 'SMO')% if we have a kernel that takes extra arguments we must define a new% kernel function handle to be passed to seqminoptif ~isempty(kfunargs)tmp_kfun = @(x,y) feval(kfun, x,y, kfunargs{:});elsetmp_kfun = kfun;end[alpha, bias] = seqminopt(training, groupIndex, ...boxconstraint, tmp_kfun, smo_opts);svIndex = find(alpha > sqrt(eps));sv = training(svIndex,:);alphaHat = groupIndex(svIndex).*alpha(svIndex);else % QP and LS both need the kernel matrix:% calculate kernel function and add additional term required% for two-norm soft margintrykx = feval(kfun,training,training,kfunargs{:});% ensure function is symmetrickx = (kx+kx')/2 + diag(1./boxconstraint);catch MEm = message('stats:svmtrain:KernelFunctionError',func2str(kfun));throw(addCause(MException(m.Identifier,'%s',getString(m)),ME));end% create HessianH =((groupIndex * groupIndex').*kx);if strcmpi(optimMethod, 'QP')if strncmpi(qp_opts.Algorithm,'inte',4)X0 = [];elseX0= ones(nPoints,1);end[alpha, ~, exitflag, output] = quadprog(H,-ones(nPoints,1),[],[],...groupIndex',0,zeros(nPoints,1), Inf *ones(nPoints,1),...X0, qp_opts);if exitflag <= 0error(message('stats:svmtrain:UnsolvableOptimization', output.message));end% The support vectors are the non-zeros of alpha.% We could also use the zero values of the Lagrangian (fifth output of% quadprog) though the method below seems to be good enough.svIndex = find(alpha > sqrt(eps));sv = training(svIndex,:);% calculate the parameters of the separating line from the support% vectors.alphaHat = groupIndex(svIndex).*alpha(svIndex);% Calculate the bias by applying the indicator function to the support% vector with largest alpha.[~,maxPos] = max(alpha);bias = groupIndex(maxPos) - sum(alphaHat.*kx(svIndex,maxPos));% an alternative method is to average the values over all support vectors% bias = mean(groupIndex(sv)' - sum(alphaHat(:,ones(1,numSVs)).*kx(sv,sv)));% An alternative way to calculate support vectors is to look for zeros of% the Lagrangian (fifth output from QUADPROG).%% [alpha,fval,output,exitflag,t] = quadprog(H,-ones(nPoints,1),[],[],...% groupIndex',0,zeros(nPoints,1),inf *ones(nPoints,1),zeros(nPoints,1),opts);%% sv = t.lower < sqrt(eps) & t.upper < sqrt(eps);else % Least-Squares% now build up compound matrix for solverA = [0 groupIndex';groupIndex,H];b = [0;ones(size(groupIndex))];x = A\b;% calculate the parameters of the separating line from the support% vectors.sv = training;bias = x(1);alphaHat = groupIndex.*x(2:end);svIndex = (1:nPoints)';end
end
svm_struct.SupportVectors = sv;
svm_struct.Alpha = alphaHat;
svm_struct.Bias = bias;
svm_struct.KernelFunction = kfun;
svm_struct.KernelFunctionArgs = kfunargs;
svm_struct.GroupNames = groupnames;
svm_struct.SupportVectorIndices = svIndex;
svm_struct.ScaleData = scaleData;
svm_struct.FigureHandles = [];
if plotflaghSV = svmplotsvs(hAxis,hLines,groupString,svm_struct);svm_struct.FigureHandles = {hAxis,hLines,hSV};
endfunction [out,f] = svmdecision_r2015a(Xnew,svm_struct)
%SVMDECISION Evaluates the SVM decision function% Copyright 2004-2012 The MathWorks, Inc.sv = svm_struct.SupportVectors;
alphaHat = svm_struct.Alpha;
bias = svm_struct.Bias;
kfun = svm_struct.KernelFunction;
kfunargs = svm_struct.KernelFunctionArgs;f = (feval(kfun,sv,Xnew,kfunargs{:})'*alphaHat(:)) + bias;
out = sign(f);
% points on the boundary are assigned to class 1
out(out==0) = 1;
function options = svmsmoset_r2015a(varargin)
%SVMSMOSET Obsolete function.
% SVMSMOSET will be removed in a future release. Use FITCSVM instead.
%
% OPTIONS = SVMSMOSET('NAME1',VALUE1,'NAME2',VALUE2,...) creates an
% options structure OPTIONS in which the named properties have the
% specified values. Any unspecified properties have default values. It is
% sufficient to type only the leading characters that uniquely identify
% the property. Case is ignored for property names.
%
% OPTIONS = SVMSMOSET(OLDOPTS,'NAME1',VALUE1,...) alters an existing
% options structure OLDOPTS.
%
% OPTIONS = SVMSMOSET(OLDOPTS,NEWOPTS) combines an existing options
% structure OLDOPTS with a new options structure NEWOPTS. Any new
% properties overwrite corresponding old properties.
%
% SVMSMOSET with no input arguments displays all property names and their
% possible values.
%
% SVMSMOSET has the following properties:
%
% TolKKT
% Tolerance with which the Karush-Kuhn-Tucker (KKT) conditions are
% checked. Default value is 1e-3.
%
% MaxIter
% Maximum number of iterations of main loop. If this number is exceeded
% before the algorithm converges then the algorithm stops and gives an
% error. Default value is 15000.
%
% Display
% Controls the level of information about the optimization iterations
% that is displayed as the algorithm runs. The value can be 'off', which
% displays nothing, 'iter', which reports every 500 iterations, and
% 'final', which reports when the algorithm finishes. Default value is
% 'off'.
%
% KKTViolationLevel
% This number specifies the fraction of alphas that are allowed to
% violate the KKT conditions. Setting this to a value greater than 0 will
% help the algorithm to converge if it is fluctuating near a good
% solution. Default value is 0.
%
% KernelCacheLimit
% This number specifies the size of the kernel matrix cache. The
% algorithm keeps a matrix with up to KernelCacheLimit * KernelCacheLimit
% double numbers in memory. Default value is 5000.
%
% Examples:
%
% opts = svmsmoset('Display','final','MaxIter',20000,...
% 'KernelCacheLimit',1000);
% alt_opts = svmsmoset(opts,'Display','iter','KKTViolationLevel',.05);
%
% See also SVMCLASSIFY, SVMTRAIN, fitcsvm.% References:
%
% [1] Cristianini, N., Shawe-Taylor, J An Introduction to Support
% Vector Machines, Cambridge University Press, Cambridge, UK. 2000.
% http://www.support-vector.net
% [2] J.C. Platt: A Fast Algorithm for Training Support Vector
% Machines, http://research.microsoft.com/users/jplatt/smo.html
% [3] R.-E. Fan, P.-H. Chen, and C.-J. Lin. Working Set Selection Using
% Second Order Information for Training SVM. Journal of Machine
% Learning Research, 6(2005), 1889-1918.
% [4] L. Bottou and C.-J. Lin. Support Vector Machine Solvers. 2006,
% available from http://www.csie.ntu.edu.tw/~cjlin/papers.html% Copyright 2006-2014 The MathWorks, Inc.%
% MaxNonBoundsIter -- may get added at a later date. Currently hardcoded
% Maximum number of iterations of the loop which tries to make the set of
% non-bound alphas (true support vectors) consistent. If this number is
% exceeded the algorithm continues with loop over the full set of alphas.
% Tuning this number can speed up the algorithm. Default value is 25.warning(message('stats:obsolete:ReplaceThisWith','svmsmoset','fitcsvm'));% Print out possible values of properties.
if (nargin == 0) && (nargout == 0)fprintf(' Display: [ off | iter | final ]\n');fprintf(' TolKKT: [ positive scalar ]\n');fprintf(' MaxIter: [ positive scalar ]\n');fprintf(' KernelCacheLimit: [ positive scalar ]\n');fprintf(' KKTViolationLevel: [ positive scalar]\n');fprintf('\n');return;
end% Create a struct of all the fields with all values set to
Options = {...'Display', 'off';'TolKKT', 1e-3;'MaxIter', 15000;'KKTViolationLevel', 0;'KernelCacheLimit', 5000;};Names = Options(:,1);
Defaults = Options(:,2);m = size(Names,1);% Combine all leading options structures o1, o2, ... in odeset(o1,o2,...).
for j = 1:moptions.(Names{j}) = Defaults{j};
end
% work through the inputs until we find a parameter name. Handle options
% structures as we go.
i = 1;
while i <= narginarg = varargin{i};if ischar(arg) % arg is an option namebreak;endif ~isempty(arg) % [] is a valid options argumentif ~isa(arg,'struct')error(message('stats:svmtrain:NoPropNameOrStruct', i));endfor j = 1:mif any(strcmp(fieldnames(arg),Names{j}))val = arg.(Names{j});elseval = [];endif ~isempty(val)options.(Names{j}) = val;endendendi = i + 1;
end% A finite state machine to parse name-value pairs.
if rem(nargin-i+1,2) ~= 0error(message('stats:svmtrain:ArgNameValueMismatch'));
end
expectval = 0; % start expecting a name, not a value
while i <= narginarg = varargin{i};if ~expectvalif ~ischar(arg)error(message('stats:svmtrain:NoPropName', i));endk = find(strncmpi(arg, Names,numel(arg)));if isempty(k)error(message('stats:svmtrain:UnknownParameterName', arg));elseif length(k)>1error(message('stats:svmtrain:AmbiguousParameterName', arg));endexpectval = 1; % we expect a value nextelseoptions.(Names{k}) = arg;expectval = 0;endi = i + 1;
endif expectvalerror(message('stats:svmtrain:NoValueForProp', arg));
end%check tolkkt其他
刚开始在网上搜索到 下载libsvm包并将其添加到Matlab toolbox中，可以继续使用svmtrain和svmclassify/svmpredict函数，尝试之后发现还是无法直接调用原来训练好的svm model，只能重新训练model
参考：
- 关于matlab2018a版本错误使用 svmclassify 分类器
- Matlab代码提示“svmtrain已删除 请改用fitcsvm”，以及svmpredict没有返回结果label和精度accuracy的解决办法
- LIBSVM – A Library for Support Vector Machines
- Old Version of LIBSVM
这篇关于用Matlab 2015a svmtrain函数训练的SVM model在2021b无法使用的解决方法的文章就介绍到这儿，希望我们推荐的文章对编程师们有所帮助！
