【Python神经网络预测】

目的&＃xff1a;预测钢铁成锭率

导入模块

from matplotlib import pyplot as plt
from sklearn.neural_network import MLPRegressor
from sklearn.preprocessing import StandardScaler
import numpy as np
import pandas as pd

读取文件

#原数据差不多是下面那样

file_path &＃61; ‘data/成锭率.csv’
df &＃61; pd.read_csv(file_path)
Y &＃61; df.iloc[:, 14]
X &＃61; df.values[:, 0:14]

标准化转换

scaler&＃61;StandardScaler()
X &＃61; scaler.fit_transform(X)

输出均值&＃xff0c;方差

print(“拟合后的均值为&＃xff1a;”, scaler.mean_)
print(“拟合后的方差&＃xff1a;”, scaler.var_)

clf &＃61; MLPRegressor(solver&＃61;‘lbfgs’, activation&＃61;‘relu’, learning_rate_init&＃61;0.001,alpha&＃61;0.001,max_iter&＃61;1000000, hidden_layer_sizes&＃61;(40,40))
clf.fit(X,Y)

测试数据

pred &＃61; clf.predict(scaler.transform([[3.66,4.55,6.27,0.24,5.68,8.75,2.35,2.86,0.123,0.536,5.145,2.48,0.55,0.49],[4.68,4.12,7.32,4.55,6.02,8.01,2.35,2.86,0.225,1.251,5.145,2.48,0.55,0.49],[6.38,7.32,8.61,4.25,6.86,3.61,1.55,1.65,0.144,1.652,1.035,1.55,0.27,0.54],
[3.26,4.55,6.23,4.56,6.55,3.29,0.62,0.86,0.429,2.409,1.035,1.75,0.46,0.53],
[6.22,6.54,8.66,4.65,4.58,3.25,1.45,1.66,0.555,0.456,0.756,1.57,0.46,0.52],
[8.05,9.12,0.26,0.12,6.08,8.87,1.26,2.31,0.552,2.548,0.185,2.15,0.48,0.51],
[8.23,7.85,4.87,6.54,6.08,9.12,1.26,3.56,0.463,0.255,1.013,2.15,0.48,0.51],
[7.56,7.56,7.47,6.25,6.55,8.88,1.26,2.31,0.552,2.456,0.185,2.15,0.48,0.51],
[6.45,8.01,8.56,6.54,6.08,8.48,1.26,3.56,0.463,0.574,1.013,2.15,0.48,0.51]]))
print(‘回归预测结果&＃xff1a;’, pred)
ypred &＃61; clf.predict(X)
print(ypred)

输出权重矩阵&＃xff0c;系数矩阵

index&＃61;0
for w in clf.coefs_:
index &＃43;&＃61; 1
print(‘第{}层网络层:’.format(index))
print(‘权重矩阵:’, w.shape)
print(‘系数矩阵&＃xff1a;’, w)

模型评价

score &＃61; clf.score(X, Y)# 相关系数
print(np.abs(df.iloc[:,14]-ypred).mean() )

画图

plt.figure()
plt.plot(np.arange(len(Y)), Y, “bo-”, label&＃61;“真实值”) # 训练数据和训练标签
plt.plot(np.arange(len(ypred )), ypred , “ro-”, label&＃61;“预测值”) # 训练数据和模型预测的标签
plt.rcParams[‘font.sans-serif’] &＃61; [‘SimHei’] # 显示中文
plt.title(f’sklearn神经网络—拟合度:{score}\n’)
plt.legend(loc&＃61;“best”)
plt.show()

完整代码

from matplotlib import pyplot as plt
from sklearn.neural_network import MLPRegressor
from sklearn.preprocessing import StandardScaler
import numpy as np
import pandas as pd
file_path &＃61; &＃39;data/成锭率.csv&＃39;
df &＃61; pd.read_csv(file_path)
Y &＃61; df.iloc[:, 14]
X &＃61; df.values[:, 0:14]
# 标准化转换
scaler&＃61;StandardScaler()
X &＃61; scaler.fit_transform(X)
print("拟合后的均值为&＃xff1a;", scaler.mean_)
print("拟合后的方差&＃xff1a;", scaler.var_)
clf &＃61; MLPRegressor(solver&＃61;&＃39;lbfgs&＃39;, activation&＃61;&＃39;relu&＃39;, learning_rate_init&＃61;0.001,alpha&＃61;0.001,max_iter&＃61;1000000, hidden_layer_sizes&＃61;(40,40))
clf.fit(X,Y)
# 测试数据
pred &＃61; clf.predict(scaler.transform([[3.66,4.55,6.27,0.24,5.68,8.75,2.35,2.86,0.123,0.536,5.145,2.48,0.55,0.49],[4.68,4.12,7.32,4.55,6.02,8.01,2.35,2.86,0.225,1.251,5.145,2.48,0.55,0.49],[6.38,7.32,8.61,4.25,6.86,3.61,1.55,1.65,0.144,1.652,1.035,1.55,0.27,0.54],
[3.26,4.55,6.23,4.56,6.55,3.29,0.62,0.86,0.429,2.409,1.035,1.75,0.46,0.53],
[6.22,6.54,8.66,4.65,4.58,3.25,1.45,1.66,0.555,0.456,0.756,1.57,0.46,0.52],
[8.05,9.12,0.26,0.12,6.08,8.87,1.26,2.31,0.552,2.548,0.185,2.15,0.48,0.51],
[8.23,7.85,4.87,6.54,6.08,9.12,1.26,3.56,0.463,0.255,1.013,2.15,0.48,0.51],
[7.56,7.56,7.47,6.25,6.55,8.88,1.26,2.31,0.552,2.456,0.185,2.15,0.48,0.51],
[6.45,8.01,8.56,6.54,6.08,8.48,1.26,3.56,0.463,0.574,1.013,2.15,0.48,0.51]]))
print(&＃39;回归预测结果&＃xff1a;&＃39;, pred)
ypred &＃61; clf.predict(X)
print(ypred)
index&＃61;0
for w in clf.coefs_:index &＃43;&＃61; 1print(&＃39;第{}层网络层:&＃39;.format(index))print(&＃39;权重矩阵:&＃39;, w.shape)print(&＃39;系数矩阵&＃xff1a;&＃39;, w)
score &＃61; clf.score(X, Y)# 模型评价
print(np.abs(df.iloc[:,14]-ypred).mean() ) # 模型评价
plt.figure()
plt.plot(np.arange(len(Y)), Y, "bo-", label&＃61;"真实值") # 训练数据和训练标签
plt.plot(np.arange(len(ypred )), ypred , "ro-", label&＃61;"预测值") # 训练数据和模型预测的标签
plt.rcParams[&＃39;font.sans-serif&＃39;] &＃61; [&＃39;SimHei&＃39;] # 显示中文
plt.title(f&＃39;sklearn神经网络---拟合度:{score}\n&＃39;)
plt.legend(loc&＃61;"best")
plt.show()