Method fit

svm_class/svm_smo.py:198–246

(self, X, Y, tol=0.00001, eps=0.01)

    return 0

  def fit(self, X, Y, tol=0.00001, eps=0.01):
    # we need these to make future predictions
    self.tol = tol
    self.eps = eps
    self.Xtrain = X
    self.Ytrain = Y
    self.N = X.shape[0]
    self.alphas = np.zeros(self.N)
    self.b = 0.
    self.errors = self._decision_function(self.Xtrain) - self.Ytrain

    # kernel matrix
    self.K = self.kernel(X, X)
    self.YY = np.outer(Y, Y)
    self.YYK = self.K * self.YY

    iter_ = 0
    numChanged = 0
    examineAll = 1
    losses = []

    while numChanged > 0 or examineAll:
      print("iter:", iter_)
      iter_ += 1
      numChanged = 0
      if examineAll:
        # loop over all training examples
        for i in range(self.alphas.shape[0]):
          examine_result = self._examine_example(i)
          numChanged += examine_result
          if examine_result:
            loss = self._loss(self.Xtrain, self.Ytrain)
            losses.append(loss)
      else:
        # loop over examples where alphas are not already at their limits
        for i in np.where((self.alphas != 0) & (self.alphas != self.C))[0]:
            examine_result = self._examine_example(i)
            numChanged += examine_result
            if examine_result:
              loss = self._loss(self.Xtrain, self.Ytrain)
              losses.append(loss)
      if examineAll == 1:
          examineAll = 0
      elif numChanged == 0:
          examineAll = 1

    plt.plot(losses)
    plt.title("loss per iteration")
    plt.show()

  def _decision_function(self, X):
    return (self.alphas * self.Ytrain).dot(self.kernel(self.Xtrain, X)) - self.b