Svm: Difference between revisions

From Eigenvector Research Documentation Wiki
Jump to navigation Jump to search
imported>Donal
imported>Donal
Line 67: Line 67:


====epsilon-SVR and nu-SVR====
====epsilon-SVR and nu-SVR====
There are two commonly used versions of SVM regression, 'epsilon-SVR' and 'nu-SVR'. The original SVM formulations for Classification (SVC) and Regression (SVR) used parameters C [0, inf) and epsilon[0, inf) to apply a penalty to the optimization for points which were not correctly separated by the classifying hyperplane or for prediction errors greater than epsilonAlternative versions of both SVM classification and regression were later developed where these penalty parameters were replaced by an alternative parameter, nu [0,1], which applies a slightly different penalty. The main motivation for the nu versions of SVM is that it has a has a more meaningful interpretation. This is because nu represents an upper bound on the fraction of training samples which are errors (misclassified, or poorly predicted) and a lower bound on the fraction of samples which are support vectors. Some users feel nu is more intuitive to use than C or epsilon.
There are two commonly used versions of SVM regression, 'epsilon-SVR' and 'nu-SVR'. The original SVM formulations for Regression (SVR) used parameters C [0, inf) and epsilon[0, inf) to apply a penalty to the optimization for points which were not correctly predictedAn alternative version of both SVM regression was later developed where the epsilon penalty parameter was replaced by an alternative parameter, nu [0,1], which applies a slightly different penalty. The main motivation for the nu versions of SVM is that it has a has a more meaningful interpretation. This is because nu represents an upper bound on the fraction of training samples which are errors (badly predicted) and a lower bound on the fraction of samples which are support vectors. Some users feel nu is more intuitive to use than C or epsilon.
C/epsilon or nu are just different versions of the penalty parameter. The same optimization problem is solved in either case. Thus it should not matter which form of SVM you use, C versus nu for classification or epsilon versus nu for regression. PLS_Toolbox uses the C and epsilon versions since these were the original formulations and are the most commonly used forms. For more details on 'nu' SVMs see [http://www.csie.ntu.edu.tw/~cjlin/papers/nusvmtutorial.pdf]
Epsilon or nu are just different versions of the penalty parameter. The same optimization problem is solved in either case. Thus it should not matter which form of SVM you use, epsilon or nu. PLS_Toolbox uses epsilon since this was the original formulation and is the most commonly used form. For more details on 'nu' SVM regression see [http://www.csie.ntu.edu.tw/~cjlin/papers/newsvr.pdf]


====SVM Parameters====
====SVM Parameters====

Revision as of 11:00, 10 October 2011

Purpose

SVM Support Vector Machine (LIBSVM) for regression. Use SVMDA for SVM classification (Svmda).

Synopsis

model = svm(x,y,options); %identifies model (calibration step).
model = svm(x,model,options); %makes predictions with a new X-block
model = svm(x,y,model,options); %performs a "test" call with a new X-block and known y-values

Description

SVM performs calibration and application of Support Vector Machine (SVM) regression models. These are non-linear models which can be used for regression or classification problems. The model consists of a number of support vectors (essentially samples selected from the calibration set) and non-linear model coefficients which define the non-linear mapping of variables in the input x-block. The model allows prediction of the continuous y-block variable (for regression problems), or the classification as passed in either the classes of the x-block or in a y-block which contains numerical classes. It is recommended that classification be done through the svmda function.

Svm is implemented using the LIBSVM package which provides both epsilon-support vector regression (epsilon-SVR) and nu-support vector regression (nu-SVR). Linear and Gaussian Radial Basis Function kernel types are supported by this function.

Note: Calling svm with no inputs starts the graphical user interface (GUI) for this analysis method.

Inputs

  • x = X-block (predictor block) class "double" or "dataset", containing numeric values,
  • y = Y-block (predicted block) class "double" or "dataset", containing numeric values,
  • model = previously generated model (when applying model to new data).

Outputs

  • model = a standard model structure model with the following fields (see MODELSTRUCT):
    • modeltype: 'SVM',
    • datasource: structure array with information about input data,
    • date: date of creation,
    • time: time of creation,
    • info: additional model information,
    • pred: 2 element cell array with
      • model predictions for each input block (when options.blockdetail='normal' x-block predictions are not saved and this will be an empty array)
    • detail: sub-structure with additional model details and results, including:
      • model.detail.svm.model: Matlab version of the libsvm svm_model (Java)
      • model.detail.svm.cvscan: Results of CV parameter scan
      • model.detail.svm.svindices: Indices of X-block samples which are support vectors.
  • pred a structure, similar to model for the new data.

Options

options = a structure array with the following fields:

  • display: [ 'off' | {'on'} ], governs level of display to command window,
  • plots [ 'none' | {'final'} ], governs level of plotting,
  • preprocessing: {[]} preprocessing structures for x block (see PREPROCESS). NOTE that y-block preprocessing is NOT used with SVMs. Any y-preprocessing will be ignored.
  • compression: [{'none'}| 'pca' | 'pls' ] type of data compression to perform on the x-block prior to calculaing or applying the SVM model. 'pca' uses a simple PCA model to compress the information. 'pls' uses either a pls or plsda model (depending on the svmtype). Compression can make the SVM more stable and less prone to overfitting.
  • compressncomp: [1] Number of latent variables (or principal components to include in the compression model.
  • blockdetails: [ {'standard'} | 'all' ], extent of predictions and residuals included in model, 'standard' = only y-block, 'all' x- and y-blocks.
  • algorithm: [ 'libsvm' ] algorithm to use. libsvm is default and currently only option.
  • kerneltype: [ 'linear' | {'rbf'} ], SVM kernel to use. 'rbf' is default.
  • svmtype: [ {'epsilon-svr'} | 'nu-svr' ] Type of SVM to apply. The default is 'epsilon-svr' for regression.
  • probabilityestimates: [0| {1} ], whether to train the SVR model for probability estimates, 0 or 1 (default 1)"
  • cvtimelimit: Set a time limit (seconds) on individual cross-validation sub-calculation when searching over supplied SVM parameter ranges for optimal parameters. Only relevant if parameter ranges are used for SVM parameters such as cost, epsilon, gamma or nu. Default is 10;
  • splits: Number of subsets to divide data into when applying n-fold cross validation. Default is 5.
  • gamma: Value(s) to use for LIBSVM kernel gamma parameter. Default is 15 values from 10^-6 to 10, spaced uniformly in log.
  • cost: Value(s) to use for LIBSVM 'c' parameter. Default is 11 values from 10^-3 to 100, spaced uniformly in log.
  • epsilon: Value(s) to use for LIBSVM 'p' parameter (epsilon in loss function). Default is the set of values [1.0, 0.1, 0.01].
  • nu: Value(s) to use for LIBSVM 'n' parameter (nu of nu-SVC, and nu-SVR). Default is the set of values [0.2, 0.5, 0.8].

Algorithm

Svm uses the LIBSVM implementation using the user-specified values for the LIBSVM parameters (see options above). See [1] for further details of these options.

The default SVM parameters cost, epsilon, nu and gamma have value ranges rather than single values. This svm function uses a search over the grid of appropriate parameters using cross-validation to select the optimal SVM parameter values and builds an SVM model using those values. This is the recommended usage. The user can avoid this grid-search by passing in single values for these parameters, however.

epsilon-SVR and nu-SVR

There are two commonly used versions of SVM regression, 'epsilon-SVR' and 'nu-SVR'. The original SVM formulations for Regression (SVR) used parameters C [0, inf) and epsilon[0, inf) to apply a penalty to the optimization for points which were not correctly predicted. An alternative version of both SVM regression was later developed where the epsilon penalty parameter was replaced by an alternative parameter, nu [0,1], which applies a slightly different penalty. The main motivation for the nu versions of SVM is that it has a has a more meaningful interpretation. This is because nu represents an upper bound on the fraction of training samples which are errors (badly predicted) and a lower bound on the fraction of samples which are support vectors. Some users feel nu is more intuitive to use than C or epsilon. Epsilon or nu are just different versions of the penalty parameter. The same optimization problem is solved in either case. Thus it should not matter which form of SVM you use, epsilon or nu. PLS_Toolbox uses epsilon since this was the original formulation and is the most commonly used form. For more details on 'nu' SVM regression see [2]

SVM Parameters

  • cost: Cost [0 ->inf] represents the penalty associated with errors larger than epsilon. Increasing cost value causes closer fitting to the calibration/training data.
  • gamma: Kernel gamma parameter controls the shape of the separating hyperplane. Increasing gamma usually increases number of support vectors.
  • epsilon: In training the regression function there is no penalty associated with points which are predicted within distance epsilon from the actual value. Decreasing epsilon forces closer fitting to the calibration/training data.
  • nu: Nu (0 -> 1] indicates a lower bound on the number of support vectors to use, given as a fraction of total calibration samples, and an upper bound on the fraction of training samples which are errors (poorly predicted).

See Also

analysis, svmda