//----------------------------------------------------------------------------
//  Copyright (C) 2004-2017 by EMGU Corporation. All rights reserved.       
//----------------------------------------------------------------------------

using System;
using Emgu.CV.CvEnum;
using Emgu.CV.ML.MlEnum;
using Emgu.CV.Structure;
using Emgu.Util;
using Emgu.CV.ML.Structure;
using System.Runtime.InteropServices;

namespace Emgu.CV.ML
{
   /// <summary>
   /// Support Vector Machine 
   /// </summary>
   public partial class SVM : UnmanagedObject, IStatModel
   {
      /// <summary>
      /// Type of SVM
      /// </summary>
      public enum SvmType
      {
         /// <summary>
         /// n-class classification (n>=2), allows imperfect separation of classes with penalty multiplier C for outliers
         /// </summary>
         CSvc = 100,
         /// <summary>
         /// n-class classification with possible imperfect separation. Parameter nu (in the range 0..1, the larger the value, the smoother the decision boundary) is used instead of C
         /// </summary>
         NuSvc = 101,
         /// <summary>
         /// one-class SVM. All the training data are from the same class, SVM builds a boundary that separates the class from the rest of the feature space
         /// </summary>
         OneClass = 102,
         /// <summary>
         /// Regression. The distance between feature vectors from the training set and the fitting hyper-plane must be less than p. For outliers the penalty multiplier C is used
         /// </summary>
         EpsSvr = 103,
         /// <summary>
         /// Regression; nu is used instead of p.
         /// </summary>
         NuSvr = 104
      }


      /// <summary>
      /// SVM kernel type
      /// </summary>
      public enum SvmKernelType
      {
         /// <summary>
         /// Custom svm kernel type
         /// </summary>
         Custom = -1,
         /// <summary>
         /// No mapping is done, linear discrimination (or regression) is done in the original feature space. It is the fastest option. d(x,y) = x y == (x,y)
         /// </summary>
         Linear = 0,
         /// <summary>
         /// polynomial kernel: d(x,y) = (gamma*(xy)+coef0)^degree
         /// </summary>
         Poly = 1,
         /// <summary>
         /// Radial-basis-function kernel; a good choice in most cases: d(x,y) = exp(-gamma*|x-y|^2)
         /// </summary>
         Rbf = 2,
         /// <summary>
         /// sigmoid function is used as a kernel: d(x,y) = tanh(gamma*(xy)+coef0)
         /// </summary>
         Sigmoid = 3,
         /// <summary>
         /// Exponential Chi2 kernel, similar to the RBF kernel
         /// </summary>
         Chi2 = 4,
         /// <summary>
         /// Histogram intersection kernel. A fast kernel. K(xi,xj)=min(xi,xj).
         /// </summary>
         Inter = 5
      }

      /// <summary>
      /// The type of SVM parameters
      /// </summary>
      public enum ParamType
      {
         /// <summary>
         /// C
         /// </summary>
         C = 0,
         /// <summary>
         /// Gamma
         /// </summary>
         Gamma = 1,
         /// <summary>
         /// P
         /// </summary>
         P = 2,
         /// <summary>
         /// NU
         /// </summary>
         Nu = 3,
         /// <summary>
         /// COEF
         /// </summary>
         Coef = 4,
         /// <summary>
         /// DEGREE
         /// </summary>
         Degree = 5
      }

      private IntPtr _statModelPtr;
      private IntPtr _algorithmPtr;

      /// <summary>
      /// Create a support Vector Machine
      /// </summary>
      public SVM()
      {
         _ptr = MlInvoke.CvSVMDefaultCreate(ref _statModelPtr, ref _algorithmPtr);
      }

      /// <summary>
      /// Release all the memory associated with the SVM
      /// </summary>
      protected override void DisposeObject()
      {
         MlInvoke.CvSVMRelease(ref _ptr);
         _statModelPtr = IntPtr.Zero;
         _algorithmPtr = IntPtr.Zero;
      }

    

      /// <summary>
      /// Get the default parameter grid for the specific SVM type
      /// </summary>
      /// <param name="type">The SVM type</param>
      /// <returns>The default parameter grid for the specific SVM type </returns>
      public static MCvParamGrid GetDefaultGrid(SVM.ParamType type)
      {
         MCvParamGrid grid = new MCvParamGrid();
         MlInvoke.CvSVMGetDefaultGrid(type, ref grid);
         return grid;
      }

      /// <summary>
      /// The method trains the SVM model automatically by choosing the optimal parameters C, gamma, p, nu, coef0, degree from CvSVMParams. By the optimality one mean that the cross-validation estimate of the test set error is minimal. 
      /// </summary>
      /// <param name="trainData">The training data.</param>
      /// <param name="kFold">Cross-validation parameter. The training set is divided into k_fold subsets, one subset being used to train the model, the others forming the test set. So, the SVM algorithm is executed k_fold times</param>
      /// <returns></returns>
      public bool TrainAuto(
         TrainData trainData,
         int kFold = 10)
      {
         return TrainAuto(
            trainData,
            kFold,
            GetDefaultGrid(ParamType.C),
            GetDefaultGrid(ParamType.Gamma),
            GetDefaultGrid(ParamType.P),
            GetDefaultGrid(ParamType.Nu),
            GetDefaultGrid(ParamType.Coef),
            GetDefaultGrid(ParamType.Degree));
      }

      /// <summary>
      /// The method trains the SVM model automatically by choosing the optimal parameters C, gamma, p, nu, coef0, degree from CvSVMParams. By the optimality one mean that the cross-validation estimate of the test set error is minimal. 
      /// </summary>
      /// <param name="trainData">The training data.</param>
      /// <param name="kFold">Cross-validation parameter. The training set is divided into k_fold subsets, one subset being used to train the model, the others forming the test set. So, the SVM algorithm is executed k_fold times</param>
      /// <param name="cGrid">cGrid</param>
      /// <param name="gammaGrid">grid for gamma</param>
      /// <param name="pGrid">grid for p</param>
      /// <param name="nuGrid">grid for nu</param>
      /// <param name="coefGrid">grid for coeff</param>
      /// <param name="degreeGrid">grid for degree</param>
      /// <param name="balanced">If true and the problem is 2-class classification then the method creates more balanced cross-validation subsets that is proportions between classes in subsets are close to such proportion in the whole train dataset.</param>
      /// <returns></returns>
      public bool TrainAuto(
         TrainData trainData,
         int kFold,
         MCvParamGrid cGrid,
         MCvParamGrid gammaGrid,
         MCvParamGrid pGrid,
         MCvParamGrid nuGrid,
         MCvParamGrid coefGrid,
         MCvParamGrid degreeGrid,
         bool balanced = false)
      {
         return MlInvoke.CvSVMTrainAuto(
            Ptr,
            trainData.Ptr,
            kFold,
            ref cGrid,
            ref gammaGrid, 
            ref pGrid, 
            ref nuGrid,
            ref coefGrid,
            ref degreeGrid,
            balanced);
      }

      /// <summary>
      /// Retrieves all the support vectors.
      /// </summary>
      /// <returns>All the support vector as floating-point matrix, where support vectors are stored as matrix rows.</returns>
      public Mat GetSupportVectors()
      {
         Mat m = new Mat();
         MlInvoke.CvSVMGetSupportVectors(_ptr, m);
         return m;
      }

      IntPtr IStatModel.StatModelPtr
      {
         get { return _statModelPtr; }
      }

      IntPtr IAlgorithm.AlgorithmPtr
      {
         get { return _algorithmPtr; }
      }
   }
}