使用TensorFlow实现SVM

较基础的SVM，后续会加上多分类以及高斯核，供大家参考。

Talk is cheap, show me the code

import tensorflow as tf
from sklearn.base import BaseEstimator, ClassifierMixin
import numpy as np

class TFSVM(BaseEstimator, ClassifierMixin):

def __init__(self,
 C = 1, kernel = 'linear',
 learning_rate = 0.01,
 training_epoch = 1000,
 display_step = 50,
 batch_size = 50,
 random_state = 42):
 #参数列表
 self.svmC = C
 self.kernel = kernel
 self.learning_rate = learning_rate
 self.training_epoch = training_epoch
 self.display_step = display_step
 self.random_state = random_state
 self.batch_size = batch_size

def reset_seed(self):
 #重置随机数
 tf.set_random_seed(self.random_state)
 np.random.seed(self.random_state)

def random_batch(self, X, y):
 #调用随机子集，实现mini-batch gradient descent
 indices = np.random.randint(1, X.shape[0], self.batch_size)
 X_batch = X[indices]
 y_batch = y[indices]
 return X_batch, y_batch

def _build_graph(self, X_train, y_train):
 #创建计算图
 self.reset_seed()

n_instances, n_inputs = X_train.shape

X = tf.placeholder(tf.float32, [None, n_inputs], name = 'X')
 y = tf.placeholder(tf.float32, [None, 1], name = 'y')

with tf.name_scope('trainable_variables'):
  #决策边界的两个变量
  W = tf.Variable(tf.truncated_normal(shape = [n_inputs, 1], stddev = 0.1), name = 'weights')
  b = tf.Variable(tf.truncated_normal([1]), name = 'bias')

with tf.name_scope('training'):
  #算法核心
  y_raw = tf.add(tf.matmul(X, W), b)
  l2_norm = tf.reduce_sum(tf.square(W))
  hinge_loss = tf.reduce_mean(tf.maximum(tf.zeros(self.batch_size, 1), tf.subtract(1., tf.multiply(y_raw, y))))
  svm_loss = tf.add(hinge_loss, tf.multiply(self.svmC, l2_norm))
  training_op = tf.train.AdamOptimizer(learning_rate = self.learning_rate).minimize(svm_loss)

with tf.name_scope('eval'):
  #正确率和预测
  prediction_class = tf.sign(y_raw)
  correct_prediction = tf.equal(y, prediction_class)
  accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))

init = tf.global_variables_initializer()

self._X = X; self._y = y
 self._loss = svm_loss; self._training_op = training_op
 self._accuracy = accuracy; self.init = init
 self._prediction_class = prediction_class
 self._W = W; self._b = b

def _get_model_params(self):
 #获取模型的参数，以便存储
 with self._graph.as_default():
  gvars = tf.get_collection(tf.GraphKeys.GLOBAL_VARIABLES)
 return {gvar.op.name: value for gvar, value in zip(gvars, self._session.run(gvars))}

def _restore_model_params(self, model_params):
 #保存模型的参数
 gvar_names = list(model_params.keys())
 assign_ops = {gvar_name: self._graph.get_operation_by_name(gvar_name + '/Assign') for gvar_name in gvar_names}
 init_values = {gvar_name: assign_op.inputs[1] for gvar_name, assign_op in assign_ops.items()}
 feed_dict = {init_values[gvar_name]: model_params[gvar_name] for gvar_name in gvar_names}
 self._session.run(assign_ops, feed_dict = feed_dict)

def fit(self, X, y, X_val = None, y_val = None):
 #fit函数，注意要输入验证集
 n_batches = X.shape[0] // self.batch_size

self._graph = tf.Graph()
 with self._graph.as_default():
  self._build_graph(X, y)

best_loss = np.infty
 best_accuracy = 0
 best_params = None
 checks_without_progress = 0
 max_checks_without_progress = 20

self._session = tf.Session(graph = self._graph)

with self._session.as_default() as sess:
  self.init.run()

for epoch in range(self.training_epoch):
   for batch_index in range(n_batches):
    X_batch, y_batch = self.random_batch(X, y)
    sess.run(self._training_op, feed_dict = {self._X:X_batch, self._y:y_batch})
   loss_val, accuracy_val = sess.run([self._loss, self._accuracy], feed_dict = {self._X: X_val, self._y: y_val})
   accuracy_train = self._accuracy.eval(feed_dict = {self._X: X_batch, self._y: y_batch})

if loss_val < best_loss:
    best_loss = loss_val
    best_params = self._get_model_params()
    checks_without_progress = 0
   else:
    checks_without_progress += 1
    if checks_without_progress > max_checks_without_progress:
     break

if accuracy_val > best_accuracy:
    best_accuracy = accuracy_val
    #best_params = self._get_model_params()

if epoch ％ self.display_step == 0:
    print('Epoch: {}\tValidaiton loss: {:.6f}\tValidation Accuracy: {:.4f}\tTraining Accuracy: {:.4f}'
     .format(epoch, loss_val, accuracy_val, accuracy_train))
  print('Best Accuracy: {:.4f}\tBest Loss: {:.6f}'.format(best_accuracy, best_loss))
  if best_params:
   self._restore_model_params(best_params)
   self._intercept = best_params['trainable_variables/weights']
   self._bias = best_params['trainable_variables/bias']
  return self

def predict(self, X):
 with self._session.as_default() as sess:
  return self._prediction_class.eval(feed_dict = {self._X: X})

def _intercept(self):
 return self._intercept

def _bias(self):
 return self._bias

实际运行效果如下（以Iris数据集为样本）：

画出决策边界来看看：

来源：https://blog.csdn.net/sdoddyjm68/article/details/79392230