原文自：http://www.csie.ntu.edu.tw/~cjlin/libsvm/faq.html#faq101

起因是不知道SVC和nu-SVC的区别，原本想瞎猜，幸好瞎猜之前搜了一下，找到了这个。

LIBSVMFAQ

Q: Some courses which have used libsvm as atool

[GoTop]

[GoTop]

[GoTop]

See thechange log.

You can download earlier versions here.

[GoTop]

Please cite the following paper:

Chih-Chung Chang and Chih-Jen Lin, LIBSVM : a library forsupport vector machines. ACM Transactions on Intelligent Systemsand Technology, 2:27:1--27:27, 2011. Software available athttp://www.csie.ntu.edu.tw/~cjlin/libsvm

The bibtex format is

@article{CC01a, author = {Chang, Chih-Chung and Lin, Chih-Jen}, title = {{LIBSVM}: A library for support vector machines}, journal = {ACM Transactions on Intelligent Systems and Technology}, volume = {2}, issue = {3}, year = {2011}, pages = {27:1--27:27}, note =  {Software available at url{http://www.csie.ntu.edu.tw/~cjlin/libsvm}}}

[GoTop]

The libsvm license ("the modified BSD license") is compatiblewith many free software licenses such as GPL. Hence, it is veryeasy to use libsvm in your software. Please check the COPYRIGHTfile in detail. Basically you need to

1. Clearly indicate that LIBSVM is used.
2. Retain the LIBSVM COPYRIGHT file in your software.

[GoTop]

Yes, see libsvmtools

[GoTop]

This usually happens if you compile the code on one machine andrun it on another which has incompatible libraries. Try torecompile the program on that machine or use static linking.

[GoTop]

Build it as a project by choosing "Win32 Project." On the otherhand, for "svm-train" and "svm-predict" you want to choose "Win32Console Project." After libsvm 2.5, you can also use the fileMakefile.win. See details in README.

If you are not using Makefile.win and see the following linkerror

LIBCMTD.lib(wwincrt0.obj) : error LNK2001: unresolved external symbol_wWinMain@16

[GoTop]

You need to open a command window and type svmtrain.exe to seeall options. Some examples are in README file.

[GoTop]

"." means every 1,000 iterations (or every #data iterations isyour #data is less than 1,000). "*" means that after iterations ofusing a smaller shrunk problem, we reset to use the whole set. Seethe implementationdocument for details.

[GoTop]

Very likely the program consumes too much memory than what theoperating system can provide. Try a smaller data and see if theprogram still crashes.

[GoTop]

The easiest way is to use Makefile.win. See details in README.Alternatively, you can use Visual C++. Here is the example usingVisual Studio .Net 2008:

1. Create a Win32 empty DLL project and set (inProject->$Project_Name Properties...->Configuration) to"Release." About how to create a new dynamic link library, pleaserefer to http://msdn2.microsoft.com/en-us/library/ms235636(VS.80).aspx
2. Add svm.cpp, svm.h to your project.
3. Add __WIN32__ and _CRT_SECURE_NO_DEPRECATE to Preprocessordefinitions (in Project->$Project_NameProperties...->C/C++->Preprocessor)
4. Set Create/Use Precompiled Header to Not Using PrecompiledHeaders (in Project->$Project_NameProperties...->C/C++->Precompiled Headers)
5. Set the path for the Modulation Definition File svm.def (inProject->$Project_Name Properties...->Linker->input
6. Build the DLL.
7. Rename the dll file to libsvm.dll and move it to the correctpath.

[GoTop]

The warning message is like

svm.cpp:2730: warning: ignoring return value of int fscanf(FILE*, const char*, ...), declared with attribute warn_unused_result

CFLAGS = -Wall -Wconversion -O3 -fPIC

CFLAGS = -Wall -Wconversion -O3 -fPIC -U_FORTIFY_SOURCE

[GoTop]

For portability, we use only features defined in ISO C89. Notethat features in ISO C99 may not be available everywhere. Even thenewest gcc lacks some features in C99 (see http://gcc.gnu.org/c99status.htmlfor details). If the situation changes in the future, we mightconsider using these newer features.

[GoTop]

libsvm uses the so called "sparse" format where zero values donot need to be stored. Hence a data with attributes

1 0 2 0

1:1 3:2

[GoTop]

Currently libsvm supports only numerical data. You may have tochange non-numerical data to numerical. For example, you can useseveral binary attributes to represent a categorical attribute.

[GoTop]

This is a controversial issue. The kernel evaluation (i.e. innerproduct) of sparse vectors is slower so the total training time canbe at least twice or three times of that using the dense format.However, we cannot support only dense format as then we CANNOThandle extremely sparse cases. Simplicity of the code is anotherconcern. Right now we decide to support the sparse format only.

[GoTop]

We assume that you have 'n' in the end of each line. So pleasepress enter in the end of your last line.

[GoTop]

The svm-train program in libsvm conducts only a simple check ofthe input data. To do a detailed check, after libsvm 2.85, you canuse the python script tools/checkdata.py. See tools/README fordetails.

[GoTop]

We don't officially support this. But, currently LIBSVM is ableto process data in the following format:

1 1:2 2:1 # your comments

[GoTop]

It depends on your data format. A simple way is to uselibsvmwrite in the libsvm matlab/octave interface. Take a CSV(comma-separated values) file in UCI machine learning repository asan example. We download SPECTF.train. Labels are in the first column. The followingsteps produce a file in the libsvm format.

matlab> SPECTF = csvread('SPECTF.train'); % read a csv filematlab> labels = SPECTF(:, 1); % labels from the 1st columnmatlab> features = SPECTF(:, 2:end); matlab> features_sparse = sparse(features); % features must be in a sparse matrixmatlab> libsvmwrite('SPECTFlibsvm.train', labels, features_sparse);

Alternatively, you can use convert.cto convert CSV format to libsvm format.

[GoTop]

obj is the optimal objective value of the dual SVM problem. rhois the bias term in the decision function sgn(w^Tx - rho). nSV andnBSV are number of support vectors and bounded support vectors(i.e., alpha_i = C). nu-svm is a somewhat equivalent form of C-SVMwhere C is replaced by nu. nu simply shows the correspondingparameter. More details are in libsvmdocument.

[GoTop]

In the model file, after parameters and other informations suchas labels , each line represents a support vector. Support vectorsare listed in the order of "labels" shown earlier. (i.e., thosefrom the first class in the "labels" list are grouped first, and soon.) If k is the total number of classes, in front of a supportvector in class j, there are k-1 coefficients y*alpha where alphaare dual solution of the following two class problems:
1 vs j, 2 vs j, ..., j-1 vs j, j vs j+1, j vs j+2, ..., j vsk
and y=1 in first j-1 coefficients, y=-1 in the remaining k-jcoefficients. For example, if there are 4 classes, the file lookslike:

+-+-+-+--------------------+|1|1|1|                    ||v|v|v|  SVs from class 1  ||2|3|4|                    |+-+-+-+--------------------+|1|2|2|                    ||v|v|v|  SVs from class 2  ||2|3|4|                    |+-+-+-+--------------------+|1|2|3|                    ||v|v|v|  SVs from class 3  ||3|3|4|                    |+-+-+-+--------------------+|1|2|3|                    ||v|v|v|  SVs from class 4  ||4|4|4|                    |+-+-+-+--------------------+

[GoTop]

We have float as the default as you can store more numbers inthe cache. In general this is good enough but for few difficultcases (e.g. C very very large) where solutions are huge numbers, itmight be possible that the numerical precision is not enough usingonly float.

[GoTop]

In general we suggest you to try the RBF kernel first. A recentresult by Keerthi and Lin ( download paperhere) shows that if RBF is used with model selection, thenthere is no need to consider the linear kernel. The kernel matrixusing sigmoid may not be positive definite and in general it'saccuracy is not better than RBF. (see the paper by Lin and Lin (download paperhere). Polynomial kernels are ok but if a high degree is used,numerical difficulties tend to happen (thinking about dth power of(<1) goes to 0 and (>1) goes to infinity).

[GoTop]

No, libsvm solves linear/nonlinear SVMs by the same way. Sometricks may save training/testing time if the linear kernel is used,so libsvm is NOT particularly efficient for linear SVM,especially when C is large and the number of data is much largerthan the number of attributes. You can either

Please also see our SVMguide on the discussion of using RBF and linear kernels.

[GoTop]

This usually happens when the data are overfitted. If attributesof your data are in large ranges, try to scale them. Then theregion of appropriate parameters may be larger. Note that there isa scale program in libsvm.

[GoTop]

Yes, you can do the following:

> svm-scale -s scaling_parameters train_data > scaled_train_data> svm-scale -r scaling_parameters test_data > scaled_test_data

[GoTop]

For the linear scaling method, if the RBF kernel is used andparameter selection is conducted, there is no difference. Assume Miand mi are respectively the maximal and minimal values of the ithattribute. Scaling to [0,1] means

                x'=(x-mi)/(Mi-mi)

                x''=2(x-mi)/(Mi-mi)-1.

                x'-y'=(x-y)/(Mi-mi), x''-y''=2(x-y)/(Mi-mi).

Though the performance is the same, the computational time maybe different. For data with many zero entries, [0,1]-scaling keepsthe sparsity of input data and hence may save the time.

[GoTop]

Try to use the model selection tool grid.py in the pythondirectory find out good parameters. To see the importance of modelselection, please see my talk: A practicalguide to support vector classification

[GoTop]

Yes, there is a -wi options. For example, if you use

> svm-train -s 0 -c 10 -w1 1 -w-1 5 data_file

the penalty for class "-1" is larger. Note that this -w optionis for C-SVC only.

[GoTop]

Basically they are the same thing but with different parameters.The range of C is from zero to infinity but nu is always between[0,1]. A nice property of nu is that it is related to the ratio ofsupport vectors and the ratio of the training error.

[GoTop]

You may want to check your data. Each training/testing data mustbe in one line. It cannot be separated. In addition, you have toremove empty lines.

[GoTop]

In theory libsvm guarantees to converge. Therefore, this meansyou are handling ill-conditioned situations (e.g. too large/smallparameters) so numerical difficulties occur.

You may get better numerical stability by replacing

typedef float Qfloat;

typedef double Qfloat;

[GoTop]

For large problems, please specify enough cache size (i.e., -m).Slow convergence may happen for some difficult cases (e.g. -c islarge). You can try to use a looser stopping tolerance with -e. Ifthat still doesn't work, you may train only a subset of the data.You can use the program subset.py in the directory "tools" toobtain a random subset.

If you have extremely large data and face this difficulty,please contact us. We will be happy to discuss possiblesolutions.

When using large -e, you may want to check if -h 0 (noshrinking) or -h 1 (shrinking) is faster. See a related questionbelow.

[GoTop]

If the number of iterations is high, then shrinking often helps.However, if the number of iterations is small (e.g., you specify alarge -e), then probably using -h 0 (no shrinking) is better. Seethe implementationdocument for details.

[GoTop]

We print out decision values for regression. For classification,we solve several binary SVMs for multi-class cases. You can obtainvalues by easily calling the subroutine svm_predict_values. Theircorresponding labels can be obtained from svm_get_labels. Detailsare in README of libsvm package.

If you are using MATLAB/OCTAVE interface, svmpredict candirectly give you decision values. Please see matlab/README fordetails.

We do not recommend the following. But if you would like to getvalues for TWO-class classification with labels +1 and -1 (note: +1and -1 but not things like 5 and 10) in the easiest way, simplyadd

                printf("%fn", dec_values[0]*model->label[0]);

              svm_predict_values(model, x, dec_values);

[GoTop]

The distance is |decision_value| / |w|. We have |w|^2 = w^Tw =alpha^T Q alpha = 2*(dual_obj + sum alpha_i). Thus in svm.cppplease find the place where we calculate the dual objective value(i.e., the subroutine Solve()) and add a statement to printw^Tw.

[GoTop]

On 32-bit machines, the maximum addressable memory is 4GB. TheLinux kernel uses 3:1 split which means user space is 3G and kernelspace is 1G. Although there are 3G user space, the maximum dynamicallocation memory is 2G. So, if you specify -m near 2G, the memorywill be exhausted. And svm-train will fail when it asks morememory. For more details, please read this article.

The easiest solution is to switch to a 64-bit machine.Otherwise, there are two ways to solve this. If your machinesupports Intel's PAE (Physical Address Extension), you can turn onthe option HIGHMEM64G in Linux kernel which uses 4G:4G split forkernel and user space. If you don't, you can try a software `tub'which can eliminate the 2G boundary for dynamic allocated memory.The `tub' is available at http://www.bitwagon.com/tub.html.

[GoTop]

For commend-line users, use the -q option:

> ./svm-train -q heart_scale

For library users, set the global variable

extern void (*svm_print_string) (const char *);

1. Declare a function to output nothing:

   void print_null(const char *s) {}

2. Assign the output function of libsvm by

   svm_print_string = &print_null;

#if 1void info(const char *fmt,...)

#if 0void info(const char *fmt,...)

[GoTop]

An example is "LIBSVM for string data" in LIBSVM Tools.

The reason why we have two functions is as follows. For the RBFkernel exp(-g |xi - xj|^2), if we calculate xi - xj first and thenthe norm square, there are 3n operations. Thus we consider exp(-g(|xi|^2 - 2dot(xi,xj) +|xj|^2)) and by calculating all |xi|^2 inthe beginning, the number of operations is reduced to 2n. This isfor the training. For prediction we cannot do this so a regularsubroutine using that 3n operations is needed. The easiest way tohave your own kernel is to put the same code in these twosubroutines by replacing any kernel.

[GoTop]

It is one-against-one. We chose it after doing the followingcomparison: C.-W. Hsu and C.-J. Lin. Acomparison of methods for multi-class support vector machines ,IEEE Transactions on Neural Networks, 13(2002), 415-425.

"1-against-the rest" is a good method whose performance iscomparable to "1-against-1." We do the latter simply because itstraining time is shorter.

[GoTop]

LIBSVM implements "one-against-one" multi-class method, so thereare k(k-1)/2 binary models, where k is the number of classes.

We can consider two ways to conduct parameter selection.

1. For any two classes of data, a parameter selection procedure isconducted. Finally, each decision function has its own optimalparameters.
2. The same parameters are used for all k(k-1)/2 binaryclassification problems. We select parameters that achieve thehighest overall performance.

Chen, Lin, and Schölkopf, Atutorial on nu-support vector machines. Applied StochasticModels in Business and Industry, 21(2005), 111-136,

they have experimentally shown that the two methods give similarperformance. Therefore, currently the parameter selection in LIBSVMtakes the second approach by considering the same parameters forall k(k-1)/2 models.

[GoTop]

Cross validation is used for selecting good parameters. Afterfinding them, you want to re-train the whole data without the -voption.

[GoTop]

Due to random partitions of the data, on different systems CVaccuracy values may be different.

[GoTop]

If you use GNU C library, the default seed 1 is considered. Thusyou always get the same result of running svm-train -v. To havedifferent seeds, you can add the following code in svm-train.c:

#include 

srand(time(0));

srand(1);

For Java, the random number generator is initialized using thetime information. So results of two CV runs are different. To fixthe seed, after version 3.1 (released in mid 2011), you can add

svm.rand.setSeed(0);

If you use CV to select parameters, it is recommended to useidentical folds under different parameters. In this case, you canconsider fixing the seed.

[GoTop]

It is extremely easy. Taking c-svc for example, to solve

min_w w^Tw/2 + C sum max(0, 1- (y_i w^Tx_i+b))^2,

only two places of svm.cpp have to be changed. First, modify thefollowing line of solve_c_svc from

    s.Solve(l, SVC_Q(*prob,*param,y), minus_ones, y,                alpha, Cp, Cn, param->eps, si, param->shrinking);

  s.Solve(l, SVC_Q(*prob,*param,y), minus_ones, y,                alpha, INF, INF, param->eps, si, param->shrinking);

     double C;

  for(int i=0;i*kernel_function)(i,i);

        this->C = param.C;        for(int i=0;i*kernel_function)(i,i)+0.5/C;

        if(i >= start && i < len)                 data[i] += 0.5/C;

For one-class svm, the modification is exactly the same. ForSVR, you don't need an if statement like the above. Instead, youonly need a simple assignment:

   data[real_i] += 0.5/C;

For large linear L2-loss SVM, please use LIBLINEAR.

[GoTop]

You have pre-specified true positive rate in mind and thensearch for parameters which achieve similar cross-validationaccuracy.

[GoTop]

This rarely happens, but few users reported the problem. Itseems that their computers for training libsvm have the VPN clientrunning. The VPN software has some bugs and causes this problem.Please try to close or disconnect the VPN client.

[GoTop]

This problem shouldn't happen after version 2.85. If you areusing earlier versions, please download the latest one.

[GoTop]

Warning: empty z range [62.5:62.5], adjusting to [61.875:63.125]Notice: cannot contour non grid data!

Nothing is wrong and please disregard the message. It is fromgnuplot when drawing the contour.

[GoTop]

This situation may occur before version 3.17. Nothing iswrong. Very likely you have two labels +1/-1 and the first instancein your data has -1. We give the following explanation.

Internally class labels are ordered by their first occurrence inthe training set. For a k-class data, internally labels are 0, ...,k-1, and each two-class SVM considers pair (i, j) with i < j.Then class i is treated as positive (+1) and j as negative (-1).For example, if the data set has labels +5/+10 and +10 appearsfirst, then internally the +5 versus +10 SVM problem has +10 aspositive (+1) and +5 as negative (-1).

By this setting, if you have labels +1 and -1, it's possiblethat internally they correspond to -1 and +1, respectively. Somenew users have been confused about this, so after version3.17, if the data set has only two labels +1 and -1, internallywe ensure +1 to be before -1. Then class +1 is always treated aspositive in the SVM problem. Note that this is for two-classdata only.

[GoTop]

Any value is ok. In this situation, what you will use is theoutput file of svm-predict, which gives predicted class labels.

[GoTop]

It is very easy if you are using GCC 4.2 or after.

In Makefile, add -fopenmp to CFLAGS.

In class SVC_Q of svm.cpp, modify the for loop of get_Q to:

#pragma omp parallel for private(j)                         for(j=start;j

In the subroutine svm_predict_values of svm.cpp, add one line tothe for loop:

#pragma omp parallel for private(i)                 for(i=0;iSV[i],model->param);

#pragma omp parallel for private(i) reduction(+:sum)                 for(i=0;il;i++)                        sum += sv_coef[i] * Kernel::k_function(x,model->SV[i],model->param);

Then rebuild the package. Kernel evaluations in training/testingwill be parallelized. An example of running this modification on an8-core machine using the data set ijcnn1:

8 cores:

%setenv OMP_NUM_THREADS 8%time svm-train -c 16 -g 4 -m 400 ijcnn127.1sec

%setenv OMP_NUM_THREADS 1%time svm-train -c 16 -g 4 -m 400 ijcnn179.8sec

export OMP_NUM_THREADS=8

For Python interface, you need to add the -lgomp linkoption:

$(CXX) -lgomp -shared -dynamiclib svm.o -o libsvm.so.$(SHVER)

For MS Windows, you need to add /openmp in CFLAGS ofMakefile.win

[GoTop]

It's very simple. Since version 3.13, you can use thefunction

void svm_get_sv_indices(const struct svm_model *model, int *sv_indices)

               model = svm_train(&prob, &param);

         int nr_sv = svm_get_nr_sv(model);                int *sv_indices = Malloc(int, nr_sv);                svm_get_sv_indices(model, sv_indices);                for (int i=0; i

If you use matlab interface, you can directly check

model.sv_indices

[GoTop]

Although sv_indices is a member of the model structure toindicate support vectors in the training set, we do not store itscontents in the model file. The model file is mainly used in thefuture for prediction, so it is basically independent fromtraining data. Thus storing sv_indices is not necessary. Usersshould find support vectors right after the training process. Seethe previous FAQ.

[GoTop]

To construct this probability model, we internally conduct across validation, which is more time consuming than a regulartraining. Hence, in general you do parameter selection firstwithout -b 1. You only use -b 1 when good parameters have beenselected. In other words, you avoid using -b 1 and -v together.

[GoTop]

There is absolutely no reason the probability outputs guaranteeyou better accuracy. The main purpose of this option is to provideyou the probability estimates, but not to boost predictionaccuracy. From our experience, after proper parameter selections,in general with and without -b have similar accuracy. Occasionallythere are some differences. It is not recommended to compare thetwo under just a fixed parameter set as more differences will beobserved.

[GoTop]

Let's just consider two-class classification here. Afterprobability information is obtained in training, we do not have

prob > = 0.5 if and only if decision value >= 0.

So predictions may be different with -b 0 and 1.

[GoTop]

For Microsoft windows, first press the "print screen" key on thekeyboard. Open "Microsoft Paint" (included in Windows) and press"ctrl-v." Then you can clip the part of picture which you want. ForX windows, you can use the program "xv" or "import" to grab thepicture of the svm-toy window.

[GoTop]

The program svm-toy assumes both attributes (i.e. x-axis andy-axis values) are in (0,1). Hence you want to scale your data tobetween a small positive number and a number less than but veryclose to 1. Moreover, class labels must be 1, 2, or 3 (not 1.0, 2.0or anything else).

[GoTop]

Taking windows/svm-toy.cpp as an example, you need to modify itand the difference from the original file is as the following: (forfive classes of data)

30,32c30<       RGB(200,0,200),<       RGB(0,160,0),<       RGB(160,0,0)--->       RGB(200,0,200)39c37< HBRUSH brush1, brush2, brush3, brush4, brush5;---> HBRUSH brush1, brush2, brush3;113,114d110<       brush4 = CreateSolidBrush(colors[7]);<       brush5 = CreateSolidBrush(colors[8]);155,157c151<       else if(v==3) return brush3;<       else if(v==4) return brush4;<       else return brush5;--->       else return brush3;325d318<         int colornum = 5;327c320<               svm_node *x_space = new svm_node[colornum * prob.l];--->               svm_node *x_space = new svm_node[3 * prob.l];333,338c326,331<                       x_space[colornum * i].index = 1;<                       x_space[colornum * i].value = q->x;<                       x_space[colornum * i + 1].index = 2;<                       x_space[colornum * i + 1].value = q->y;<                       x_space[colornum * i + 2].index = -1;<                       prob.x[i] = &x_space[colornum * i];--->                       x_space[3 * i].index = 1;>                       x_space[3 * i].value = q->x;>                       x_space[3 * i + 1].index = 2;>                       x_space[3 * i + 1].value = q->y;>                       x_space[3 * i + 2].index = -1;>                       prob.x[i] = &x_space[3 * i];397c390<                               if(current_value > 5) current_value = 1;--->                               if(current_value > 3) current_value = 1;

[GoTop]

They are the same thing. We just rewrote the C++ code inJava.

[GoTop]

This depends on the VM you used. We have seen good VM whichleads the Java version to be quite competitive with the C++ code.(though still slower)

[GoTop]

You should try to increase the maximum Java heap size. Forexample,

java -Xmx2048m -classpath libsvm.jar svm_train ...

[GoTop]

Unlike C, Java does not have a preprocessor built-in. However,we need some macros (see first 3 lines of svm.m4).

[GoTop]

Yes, here are some examples:

$ export CLASSPATH=$CLASSPATH:~/libsvm-2.91/java/libsvm.jar$ ./jythonJython 2.1a3 on java1.3.0 (JIT: jitc)Type "copyright", "credits" or "license" for more information.>>> from libsvm import *>>> dir()['__doc__', '__name__', 'svm', 'svm_model', 'svm_node', 'svm_parameter','svm_problem']>>> x1 = [svm_node(index=1,value=1)]>>> x2 = [svm_node(index=1,value=-1)]>>> param = svm_parameter(svm_type=0,kernel_type=2,gamma=1,cache_size=40,eps=0.001,C=1,nr_weight=0,shrinking=1)>>> prob = svm_problem(l=2,y=[1,-1],x=[x1,x2])>>> model = svm.svm_train(prob,param)*optimization finished, #iter = 1nu = 1.0obj = -1.018315639346838, rho = 0.0nSV = 2, nBSV = 2Total nSV = 2>>> svm.svm_predict(model,x1)1.0>>> svm.svm_predict(model,x2)-1.0>>> svm.svm_save_model("test.model",model)

[GoTop]

Your compiler version may not be supported/compatible forMATLAB. Please check thisMATLAB page first and then specify the version number. Forexample, if g++ X.Y is supported, replace

CXX = g++

CXX = g++-X.Y

[GoTop]

Please make sure that you use the -largeArrayDims option inmake.m. For example,

mex -largeArrayDims -O -c svm.cpp

[GoTop]

It is extremely easy to do scaling under MATLAB. The followingone-line code scale each feature to the range of [0,1]:

(data - repmat(min(data,[],1),size(data,1),1))*spdiags(1./(max(data,[],1)-min(data,[],1))',0,size(data,2),size(data,2))

[GoTop]

One can do this by a simple loop. See the following example:

bestcv = 0;for log2c = -1:3,  for log2g = -4:1,    cmd = ['-v 5 -c ', num2str(2^log2c), ' -g ', num2str(2^log2g)];    cv = svmtrain(heart_scale_label, heart_scale_inst, cmd);    if (cv >= bestcv),      bestcv = cv; bestc = 2^log2c; bestg = 2^log2g;    end    fprintf('%g %g %g (best c=%g, g=%g, rate=%g)n', log2c, log2g, cv, bestc, bestg, bestcv);  endend

[GoTop]

Fabrizio Lacalandra of University of Pisa reported this issue.It seems the problem is caused by the screen output. If you disablethe info function using

#if 0,

[GoTop]

In Makefile, you need to add -fopenmp to CFLAGS and -lgomp toMEX_OPTION. For Octave, you need the same modification.

However, a minor problem is that the number of threads cannot bespecified in MATLAB. We tried Version 7.12 (R2011a) andgcc-4.6.1.

% export OMP_NUM_THREADS=4; matlab>> setenv('OMP_NUM_THREADS', '1');

#pragma omp parallel  for private(i) num_threads(4)

[GoTop]

Let's start from the binary class and assume you have two labels-1 and +1. After obtaining the model from calling svmtrain, do thefollowing to have w and b:

w = model.SVs' * model.sv_coef;b = -model.rho;if model.Label(1) == -1  w = -w;  b = -b;end

For multi-class SVM, we illustrate the setting in the followingexample of running the iris data, which have 3 classes

  > [y, x] = libsvmread('../../htdocs/libsvmtools/datasets/multiclass/iris.scale');> m = svmtrain(y, x, '-t 0')m =     Parameters: [5x1 double]      nr_class: 3       totalSV: 42           rho: [3x1 double]         Label: [3x1 double]         ProbA: []         ProbB: []           nSV: [3x1 double]       sv_coef: [42x2 double]           SVs: [42x4 double]

+-+-+--------------------+|1|1|                    ||v|v|  SVs from class 1  ||2|3|                    |+-+-+--------------------+|1|2|                    ||v|v|  SVs from class 2  ||2|3|                    |+-+-+--------------------+|1|2|                    ||v|v|  SVs from class 3  ||3|3|                    |+-+-+--------------------+

  > m.nSVans =     3    21    18

  > coef = [m.sv_coef(1:3,2); m.sv_coef(25:42,1)];

  > SVs = [m.SVs(1:3,:); m.SVs(25:42,:)];

> w = SVs'*coef;

> m.rhoans =    1.1465    0.3682   -1.9969> b = -m.rho(2);

[GoTop]

Yes, after libsvm 2.86, the matlab interface works on OCTAVE aswell. Please use make.m by typing

>> make 

[GoTop]

The easiest way is to rename the svmtrain binary file (e.g.,svmtrain.mexw32 on 32-bit windows) to a different name (e.g.,svmtrain2.mexw32).

[GoTop]

The error usually happens when there are missing runtimecomponents such as MSVCR100.dll on your Windows platform. You canuse tools such as Dependency Walker to findmissing library files.

For example, if the pre-built MEX files are compiled by VisualC++ 2010, you must have installed Microsoft Visual C++Redistributable Package 2010 (vcredist_x86.exe). You can easilyfind the freely available file from Microsoft's web site.

For 64bit Windows, the situation is similar. If the pre-builtfiles are by Visual C++ 2008, then you must have Microsoft VisualC++ Redistributable Package 2008 (vcredist_x64.exe).

[GoTop]

Please use code in the following directory.The following example shows how to train and test the problem dna(trainingand testing).

Load, train and predict data:

[trainY trainX] = libsvmread('./dna.scale');[testY testX] = libsvmread('./dna.scale.t');model = ovrtrain(trainY, trainX, '-c 8 -g 4');[pred ac decv] = ovrpredict(testY, testX, model);fprintf('Accuracy = %g%%n', ac * 100);

bestcv = 0;for log2c = -1:2:3,  for log2g = -4:2:1,    cmd = ['-q -c ', num2str(2^log2c), ' -g ', num2str(2^log2g)];    cv = get_cv_ac(trainY, trainX, cmd, 3);    if (cv >= bestcv),      bestcv = cv; bestc = 2^log2c; bestg = 2^log2g;    end    fprintf('%g %g %g (best c=%g, g=%g, rate=%g)n', log2c, log2g, cv, bestc, bestg, bestcv);  endend

爱华网本文地址 » http://www.aihuau.com/a/25101015/272444.html