TensorFlow如何实现线性支持向量机SVM

import numpy as npimport tensorflow as tffrom sklearn import datasetsx_data = tf.placeholder(shape=[None, 2], dtype=tf.float32)y_target = tf.placeholder(shape=[None, 1], dtype=tf.float32)# 获得batch大小的数据def gen_data(batch_size):    iris = datasets.load_iris()    iris_X = np.array([[x[0], x[3]] for x in iris.data])    iris_y = np.array([1 if y == 0 else -1 for y in iris.target])    train_indices = np.random.choice(len(iris_X),                 int(round(len(iris_X) * 0.8)), replace=False)    train_x = iris_X[train_indices]    train_y = iris_y[train_indices]    rand_index = np.random.choice(len(train_x), size=batch_size)    batch_train_x = train_x[rand_index]    batch_train_y = np.transpose([train_y[rand_index]])    test_indices = np.array(        list(set(range(len(iris_X))) - set(train_indices)))    test_x = iris_X[test_indices]    test_y = iris_y[test_indices]    return batch_train_x, batch_train_y, test_x, test_y# 定义模型def svm():    A = tf.Variable(tf.random_normal(shape=[2, 1]))    b = tf.Variable(tf.random_normal(shape=[1, 1]))    model_output = tf.subtract(tf.matmul(x_data, A), b)    l2_norm = tf.reduce_sum(tf.square(A))    alpha = tf.constant([0.01])    classification_term = tf.reduce_mean(tf.maximum(0.,          tf.subtract(1., tf.multiply(model_output, y_target))))    loss = tf.add(classification_term, tf.multiply(alpha, l2_norm))    my_opt = tf.train.GradientDescentOptimizer(0.01)    train_step = my_opt.minimize(loss)    return model_output, loss, train_stepdef train(sess, batch_size):    print("# Training loop")    for i in range(100):        x_vals_train, y_vals_train,\        x_vals_test, y_vals_test = gen_data(batch_size)        model_output, loss, train_step = svm()        init = tf.global_variables_initializer()        sess.run(init)        prediction = tf.sign(model_output)        accuracy = tf.reduce_mean(tf.cast(            tf.equal(prediction, y_target), tf.float32))        sess.run(train_step, feed_dict=                                   {                                    x_data: x_vals_train,                        y_target: y_vals_train                                    })        train_loss = sess.run(loss, feed_dict=                                           {                                           x_data: x_vals_train,                              y_target: y_vals_train                                           })        train_acc = sess.run(accuracy, feed_dict=                                            {                                            x_data: x_vals_train,                              y_target: y_vals_train                                            })        test_acc = sess.run(accuracy, feed_dict=                                        {                                  x_data: x_vals_test,                        y_target: np.transpose([y_vals_test])                                        })        if i % 10 == 1:            print("train loss: {:.6f}, train accuracy : {:.6f}".                  format(train_loss[0], train_acc))            print            print("test accuracy : {:.6f}".format(test_acc))            print("- * - "*15)def main(_):    with tf.Session() as sess:        train(sess, batch_size=16)if __name__ == "__main__":    tf.app.run()