以下是本文所用数据~~~
一、直接上手撸代码
import pandas as pdimport numpy as npimport mathdata = pd.read_excel('上证指数.xls')data = np.array(data.iloc[3:-1,1:])e = 1ita = 0.0035#sigmoiddef sigmoid(x):return 1/(1+np.exp(-x))#sigmoid导函数def sigmoid_derivation(x):return np.exp(x)/(np.exp(x)+1)**2#将整个矩阵归一化max1 = data.max(axis = 0)min1 = data.min(axis = 0)t = np.empty((len(data),4))for i in range(4):t[:,i] = (data[:,i]-min1[i])/(max1[i]-min1[i])#截取训练集和测试集，这里使用了最后一天为测试集train = t[:-1,:-1]#训练集train_result = t[:-1,-1]#训练结果test = t[-1,:-1]#测试集test_result = t[-1,-1]#测试结果#初始化权重和偏置值w2 = np.random.random((3,3))b2 = np.random.random((3,1))w3 = np.random.random((1,3))b3 = np.random.random((1,1))i = 0while i<len(train):#前向计算z2 = w2.dot(train[i].T) + b2.Ta2 = sigmoid(z2)z3 = w3.dot(a2.T) + b3a3 = z3#计算误差ex = train_result[i] - a3#反向计算delta3 = -((ex)*(z3))delta2 = w3.T.dot(delta3)*sigmoid_derivation(z2)dw3 = delta3.dot(a2)db3 = delta3dw2 = delta2.dot(train[i])db2 = delta3#更新矩阵w3 = w3 + ita*(-dw3)b3 = b3 + ita*(-db3)w2 = w2 + ita*(-dw2)b2 = b2 + ita*(-db2)#误差均值e = (e + ex)/i#迭代次数i = i + 1print('w2矩阵为：',w2)print('w3矩阵为：',w3)print('b2矩阵为：',b2)print('b3矩阵为：',b3)#计算预测值z2 = w2.dot(test.T) + b2.Ta2 = sigmoid(z2)z3 = w3.dot(a2.T) + b3a3 = z3#返归一化print('预测值为：',a3*(max1[3]-min1[3])+min1[3])print('真实数据为：{}'.format(test_result*(max1[3]-min1[3])+min1[3]))print('误差为：{}'.format((a3*(max1[3]-min1[3])+min1[3]) - (test_result*(max1[3]-min1[3])+min1[3])))
输出结果为：
二、调库代码
import pandas as pdimport numpy as npimport mathfrom sklearn.neural_network import MLPRegressor,MLPClassifierdata = pd.read_excel('上证指数.xls')data = np.array(data.iloc[3:-1,1:])data1 = np.array(data)X = data1[:, 0:2]Y = data1[:, 2]max1 = data.max(axis = 0)min1 = data.min(axis = 0)t = np.empty((len(data),4))for i in range(4):t[:,i] = (data[:,i]-min1[i])/(max1[i]-min1[i])#截取训练集和测试集，这里使用了最后一天为测试集train = t[:-1,:-1]#训练集train_result = t[:-1,-1]#训练结果test = t[-1,:-1]#测试集test_result = t[-1,-1]#测试结果 clf = MLPRegressor(hidden_layer_sizes=(4,1), activation='relu',solver='adam',alpha=0.0001,max_iter=2000)clf = clf.fit(train,train_result)pred = clf.predict(test.reshape(1, -1))print('回归预测结果：', pred)print('预测值为：',pred*(max1[3]-min1[3])+min1[3])print('真实数据为：{}'.format(test_result*(max1[3]-min1[3])+min1[3]))print('误差为：{}'.format((pred*(max1[3]-min1[3])+min1[3]) - (test_result*(max1[3]-min1[3])+min1[3])))
输出结果为：
