#coding:utf-8
 import numpyas np
 import pandas as pd
 import matplotlib.pyplotas plt
 import tensorflowas tf
 from sklearn.model_selection import train_test_split
 from sklearnimport preprocessing
 
 plt.rcParams['font.sans-serif'] = ['SimHei']
 plt.rcParams['axes.unicode_minus'] = False
 
 df= pd.read_csv('iris.csv', delimiter=',')
 #对类别进行数值化处理
 le = preprocessing.LabelEncoder()
 df['Cluster'] = le.fit_transform(df['Species'])
 x = df[['SepalLengthCm','SepalWidthCm','PetalLengthCm','PetalWidthCm']]
 y = df[['Cluster']]
 sess = tf.Session()
 seed = 2
 tf.set_random_seed(seed)
 np.random.seed(seed)
 #创建训练集与测试集的划分
 x_train, x_test,y_train, y_test=train_test_split(x, y, train_size=0.8, test_size=0.2)
 # 添加占位符，四个输入
 x_data = tf.placeholder(shape=[None, 4], dtype=tf.float32)

# 添加占位符，一个输出
 y_target = tf.placeholder(shape=[None, 1], dtype=tf.float32)
 # 定义如何添加一个隐藏层的函数
 def add_layer(input_layer, input_num, output_num):
     weights =  tf.Variable(tf.random_normal(shape=[input_num,  output_num]))
     biase =  tf.Variable(tf.random_normal(shape=[output_num]))
     hidden_output = tf.nn.relu(tf.add(tf.matmul(input_layer, weights),  biase))
     return hidden_output
 # 定义三个隐藏层对应的结点个数
 hidden_layer_nodes = [10,8,10]
 hidden_output = add_layer(x_data, 4, hidden_layer_nodes[0])

# 循环添加三个隐藏层
 for i in range(len(hidden_layer_nodes[:-1])):
     hidden_output =  add_layer(hidden_output, hidden_layer_nodes[i],hidden_layer_nodes[i + 1])
 final_output = add_layer(hidden_output,hidden_layer_nodes[-1],1)
 # 定义损失函数,使得误差最小
 loss = tf.reduce_mean(tf.square(y_target - final_output))
 # 设置学习率来调整每一步更新的大小
 my_opt = tf.train.GradientDescentOptimizer(learning_rate=0.00004)
 # 优化目标：最小化损失函数
 train_step = my_opt.minimize(loss)
 init = tf.global_variables_initializer()
 sess.run(init)
 loss_vec = [] #训练操失
 test_loss = [] #测试操失
 #训练次数
 for i in range(10000):
     #训练
     sess.run(train_step,  feed_dict={x_data:x_train,  y_target:y_train})
     #训练数据评估模型
     temp_loss =  sess.run(loss, feed_dict=  {x_data:x_train, y_target:y_train})
     loss_vec.append(np.sqrt(temp_loss))
     #测试数据评估模型
     test_temp_loss =  sess.run(loss, feed_dict=  {x_data:x_test, y_target:y_test})
     test_loss.append(np.sqrt(test_temp_loss))
     if (i+1)%1000 == 0:
         print('Generation:' + str(i+1) + '.Loss = ' + str(temp_loss))
 
 test_preds = [np.round(item,0) for item in sess.run(final_output,feed_dict={x_data:x_test})]
 train_preds = [np.round(item,0) for item in sess.run(final_output,feed_dict={x_data:x_train})]
 y_test = [i for i in y_test['Cluster']]
 y_train = [i for i in y_train['Cluster']]
 
 test_acc = np.mean([i==j for i, j in zip(test_preds, y_test)]) * 100
 train_acc =  np.mean([i==j for i, j in zip(train_preds, y_train)]) * 100
 print('训练数据预测精确度:{}'.format(train_acc))
 print('测试数据训练精确度:{}'.format(test_acc))
 
 plt.plot(loss_vec, 'k-', label ='训练损失')
 plt.plot(test_loss,  'r--', label ='测试损失')
 plt.title('损失')
 plt.xlabel('迭代次数')
 plt.ylabel('损失')
 plt.legend(loc='upper right')
 plt.show()