设计思路

输出结果

w_target.shape:  (3,) [ 1.17881511 -5.13265596 -6.55556511]

Pre_Logistic_function

<class 'function'>

Product_x_function

[1.         0.10262954 0.43893794]

data_x

(300, 3) [[ 1.         -0.15378708  0.9615284 ]

[ 1.          0.36965948 -0.0381362 ]

[ 1.         -0.21576496 -0.31364397]

[ 1.          0.45809941 -0.12285551]

[ 1.         -0.88064421 -0.20391149]]

核心代码

def gradient_descent(data_x, data_y, w_h=None, eta=1.0, max_iterations=10000, epsilon=0.001):

   if w_h == None:

       w_h = np.array([0.0 for i in range(data_x.shape[1])])

   w_h_i = [np.copy(w_h)]  

   

   for i in range(max_iterations):

       subset_indices = range(data_x.shape[0])

       grad_E_in = np.mean(np.tile(- data_y[subset_indices] /

                                       ( 1.0 + np.exp(data_y[subset_indices] * w_h.dot(data_x[subset_indices].T)) ),

                                   (data_x.shape[1], 1)).T *   data_x[subset_indices], axis=0)

       w_h -= eta * grad_E_in

       w_h_i.append(np.copy(w_h))

       if np.linalg.norm(grad_E_in) <= np.linalg.norm(w_h) * epsilon:

           break

   return np.array(w_h_i)

LoR = linear_model.LogisticRegression()

LoR.fit(data_x,data_y)

y_train=LoR.predict(data_x)

LoRpp_function = lambda z: LoR.predict_proba(z)[:,0]

BG_Grid_BaseLoR = apply_to_fill(z_grid, LoRpp_function)

full_N_fig = plot_dataset_and_hypothesis(3,data_x, data_y, xy_1, xy_2, BG_Grid_BaseLoR,title=r'LoR: Hypothesis, $N={:}$'.format(N))

SVM_Linear = svm.SVC(kernel='linear')

SVM_Linear.fit(data_x,data_y)

SVM_LinearPre_function = lambda z: SVM_Linear.predict(z)

BG_Grid_BaseSVM_Linear = apply_to_fill(z_grid, SVM_LinearPre_function)

full_N_fig = plot_dataset_and_hypothesis(5,data_x, data_y, xy_1, xy_2, BG_Grid_BaseSVM_Linear, title=r'SVM_Linear: Hypothesis, $N={:}$'.format(N))