预测:使用SVR模块对未来事件进行时间序列预测 [英] Prediction: Time-series prediction of future events using SVR module
问题描述
我想使用scikit-learn的SVR模块对未来事件进行时间序列预测.这是我尝试使用的源代码:
I want to perform time-series prediction of future events using SVR module from scikit-learn. Here is my source code I am trying to work with:
import csv
import numpy as np
from sklearn.svm import SVR
import matplotlib.pyplot as plt
plt.switch_backend('newbackend')
seq_num=[]
win=[]
def get_data(filename):
with open(filename, 'r') as csvfile:
csvFileReader = csv.reader(csvfile)
next(csvFileReader) # skipping column names
for row in csvFileReader:
seq_num.append(int(row[0])
win.append(int(row[6]))
return
def predict_win(X, y, x):
win = np.reshape(X,(len(X), 1))
svr_lin = SVR(kernel= 'linear', C= 1e3)
svr_poly = SVR(kernel= 'poly', C= 1e3, degree= 2)
svr_rbf = SVR(kernel= 'rbf', C= 1e3, gamma= 0.1)
svr_rbf.fit(X, y)
svr_lin.fit(X, y)
svr_poly.fit(X, y)
plt.scatter(X, y, color= 'black', label= 'Data')
plt.plot(y, svr_rbf.predict(X), color= 'red', label= 'RBF model')
plt.plot(y,svr_lin.predict(X), color= 'green', label= 'Linear model')
plt.plot(y,svr_poly.predict(X), color= 'blue', label= 'Polynomial model')
plt.xlabel('X, other features')
plt.ylabel('win')
plt.title('Support Vector Regression')
plt.legend()
plt.show()
return svr_rbf.predict(x)[0], svr_lin.predict(x)[0], svr_poly.predict(x)[0]
get_data('net_data.csv')
predicted_win = predict_win(X, y, 29)
我的数据集非常庞大，因此结尾部分包含了我的csv数据集的一部分.我对第七栏感兴趣.我想预测第7列中的值何时增加或减少.是否可以仅查看第7列并进行时间序列预测?任何帮助将不胜感激?谢谢！
My dataset is very huge and so a portion of my csv dataset is included at the end. I am interested in the 7th column. I wanted to predict when the values in the 7th column increase or when it decreases. Is it possible to look into the 7th column ONLY and do the time-series prediction? Any help with this would be so much appreciated? Thanks!
0.007804347,10.0.0.11:49438,10.0.12.12:5001,32,3796291040,3796277984,10,2147483647,28960,3034,29312
0.007856739,10.0.0.11:49438,10.0.12.12:5001,32,3796293936,3796278008,11,2147483647,29056,2999,29312
0.010605189,10.0.0.11:49438,10.0.12.12:5001,32,3796320000,3796291040,20,2147483647,55040,2969,29312
0.010850907,10.0.0.11:49438,10.0.12.12:5001,32,3796348960,3796305520,30,2147483647,84096,2946,29312
0.013598458,10.0.0.11:49438,10.0.12.12:5001,32,3796377920,3796320000,40,2147483647,113024,2951,29312
0.01368011,10.0.0.11:49438,10.0.12.12:5001,32,3796434392,3796348960,60,2147483647,170880,2956,29312
0.015104265,10.0.0.11:49438,10.0.12.12:5001,32,3796434392,3796363440,70,2147483647,199936,2940,29312
0.016406964,10.0.0.11:49438,10.0.12.12:5001,32,3796490864,3796377920,80,2147483647,220160,2943,29312
0.016465876,10.0.0.11:49438,10.0.12.12:5001,32,3796537200,3796432944,81,80,330240,2925,29312
0.018355321,10.0.0.11:49438,10.0.12.12:5001,32,3796547336,3796434392,81,80,333056,2914,29312
0.020171945,10.0.0.11:49438,10.0.12.12:5001,32,3796603808,3796490864,83,80,382336,2956,29312
0.237314523,10.0.0.11:49438,10.0.12.12:5001,32,3810417728,3809658976,529,396,1775360,7109,29312
0.237409075,10.0.0.11:49438,10.0.12.12:5001,44,3810417728,3809700968,530,397,1859328,7381,29312
0.237486647,10.0.0.11:49438,10.0.12.12:5001,44,3810417728,3809700968,371,371,1960704,7365,29312
0.237807596,10.0.0.11:49438,10.0.12.12:5001,44,3810417728,3809700968,371,371,1980928,7362,29312
0.237989588,10.0.0.11:49438,10.0.12.12:5001,44,3810417728,3809700968,371,371,1989632,7400,29312
0.259123971,10.0.0.11:49438,10.0.12.12:5001,32,3811590608,3811251776,261,260,2267648,5885,29312
0.259174008,10.0.0.11:49438,10.0.12.12:5001,32,3811655768,3811289424,261,260,2267648,5918,29312
0.262546461,10.0.0.11:49438,10.0.12.12:5001,32,3811720928,3811354584,261,260,2267648,5823,29312
推荐答案
好的，下面的svm函数有问题:
Ok, the svm function below has a problem:
第二行win = ...未使用，将导致错误.删除它.
The second line, win = ... is unused, and will cause an error. Delete it.
def predict_win(X, y, x):
win = np.reshape(X,(len(X), 1)) # <----This line
svr_lin = SVR(kernel= 'linear', C= 1e3)
svr_poly = SVR(kernel= 'poly', C= 1e3, degree= 2)
svr_rbf = SVR(kernel= 'rbf', C= 1e3, gamma= 0.1)
svr_rbf.fit(X, y)
svr_lin.fit(X, y)
svr_poly.fit(X, y)
plt.scatter(X, y, color= 'black', label= 'Data')
plt.plot(y, svr_rbf.predict(X), color= 'red', label= 'RBF model')
plt.plot(y,svr_lin.predict(X), color= 'green', label= 'Linear model')
plt.plot(y,svr_poly.predict(X), color= 'blue', label= 'Polynomial model')
plt.xlabel('X, other features')
plt.ylabel('win')
plt.title('Support Vector Regression')
plt.legend()
plt.show()
return svr_rbf.predict(x)[0], svr_lin.predict(x)[0], svr_poly.predict(x)[0]
第二，我不知道为什么有读取CSV的完整功能.忽略它并使用熊猫.这是一个有效的示例代码:
Second, I don't know why there is an entire function for reading a csv. Ignore it and use pandas. Here is a sample code that will work:
from sklearn import svm
import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
def predict_win(X,y,x):
svr_lin = svm.SVR(kernel='linear',C=1e3)
svr_poly = svm.SVR(kernel='poly',C=1e3, degree=2)
svr_rbf = svm.SVR(kernel='rbf',C=1e3,gamma=0.1)
svr_rbf.fit(X,y)
svr_lin.fit(X,y)
svr_poly.fit(X,y)
plt.plot(y,svr_rbf.predict(X),color='red',label='RBF model')
plt.plot(y,svr_lin.predict(X),color='green',label='Linear model')
plt.plot(y,svr_poly.predict(X),color='blue', label='Polynomial model')
plt.xlabel('X, other features')
plt.ylabel('win')
plt.title('Support Vector Regression')
plt.legend()
plt.show()
return [svr_rbf.predict(x)[0],svr_lin.predict(x)[0],svr_poly.predict(x)[0]]
df = pd.read_csv('data.csv')
data_np_array = df.values
y = np.ndarray.copy(data_np_array[:,6])
Xleft = np.ndarray.copy(data_np_array[:,:6])
Xright = np.ndarray.copy(data_np_array[:,7:])
X = np.hstack((Xleft,Xright))
x0 = np.ndarray.copy(X[0,:])
xp = predict_win(X,y,x0)
percent_off = [min(data_np_array[0,2],prediction)/max(data_np_array[0,2],prediction) for prediction in xp]
中间步骤是清理导入的数据，将其从数据帧转换为numpy数组，将第7列复制为适合的回归值，将其从训练数据中删除，然后重新构建新的数组在安装SVR之前.
The intermediate steps, where you clean up the imported data, turn it from a dataframe to a numpy array, copy your 7th column as the regression to fit, delete it from your training data, and rebuild a new array must be done before fitting to the SVR.
df = pd.read_csv('data.csv')
data_np_array = df.values
y = np.ndarray.copy(data_np_array[:,6])
Xleft = np.ndarray.copy(data_np_array[:,:6])
Xright = np.ndarray.copy(data_np_array[:,7:])
X = np.hstack((Xleft,Xright))
让我知道这些方法是否有效.我只是从上面的数据表中提取了几行.
Let me know if these worked. I just took a few lines from your data table above.
这篇关于预测:使用SVR模块对未来事件进行时间序列预测的文章就介绍到这了，希望我们推荐的答案对大家有所帮助，也希望大家多多支持IT屋！
