这篇博客是上一篇python-arima模型statsmodels库实现的续集,上一篇采用的statsmodels版本应该要高一点,如果使用低版本的statsmodels代码会有bug,这一篇则是针对statsmodels-0.9.0版本的代码。
代码如下:
#coding=gbk
import numpy as np
import pandas as pd
import os
from numpy import NaN
from numpy import nan
import matplotlib.pyplot as plt
import statsmodels.api as sm #acf,pacf图
from statsmodels.tsa.stattools import adfuller #adf检验
from pandas.plotting import autocorrelation_plot
from statsmodels.tsa.arima_model import ARIMA
from statsmodels.stats.diagnostic import acorr_ljungbox
import statsmodels.api as sm
import matplotlib as mpl
path="E:/data/china_data.xlsx"
# 为了控制计算量,我们限制AR最大阶不超过6,MA最大阶不超过4。
plt.style.use('fivethirtyeight')
mpl.rcParams['font.sans-serif'] = ['SimHei']
mpl.rcParams['font.serif'] = ['SimHei']
mpl.rcParams['axes.unicode_minus'] = False
df=pd.read_excel(path)
#print(df)
#help(df)
#for index, row in df.iterrows():
df=df.replace(NaN, "null")
# print(index, row)
print(df)
def f(column):
r=0
inde1=0
index2=len(column)-1
for i in range(len(column)):
# print(column[len(column)-i-1])
if column[len(column)-i-1] is "null" and r==1:
index2=i
return index1,index2
if column[len(column)-i-1]!= "null" and r==0:
index1=i
r=1
return index1,index2
#df['时间(年)']=pd.to_datetime(df['时间(年)'])
print(df.columns)
print(df[df.columns[0]])
indexz=df.columns[0]
def adf_test(data):#小于0.05则是平稳序列
# print("data:",data.values)
data_z=np.array(list(data.values))
#print(data_z.reshape(-1,))
t = adfuller(data_z.reshape(-1,))
print("p-value:",t[1])
def box_pierce_test(data):#小于0.05,不是白噪声序列
print(acorr_ljungbox(data, lags=1))
def stability_judgment(data):
fig = plt.figure(figsize=(12,8))
ax1=fig.add_subplot(211)
fig = sm.graphics.tsa.plot_acf(data,lags=5,ax=ax1)
ax2 = fig.add_subplot(212)
fig = sm.graphics.tsa.plot_pacf(data,lags=5,ax=ax2)
plt.show()
def model_fit(data,df,index,length,index1,index2):
data_diff=df[["时间(年)",index]][length-index2:length-index1]
# sm.tsa.arma_order_select_ic(data_diff,max_ar=6,max_ma=4,ic='aic')['aic_min_order'] # AIC
#对模型进行定阶
pmax = int(len(data) / 10) #一般阶数不超过 length /10
qmax = int(len(data) / 10)
if pmax>4:pmax=6
if qmax>4:qmax=4
bic_matrix = []
print("data",data)
# help(sm.tsa.arima.ARIMA)
for p in range(pmax +1):
temp= []
for q in range(qmax+1):
try:
# ARIMA(train_data, order=(1,1,1))
# print(sm.tsa.arima.ARIMA(data,order=(p,1,q)).fit())
temp.append(sm.tsa.ARIMA(data,order=(p,1,q)).fit().bic)
# print(temp)
except:
temp.append(None)
# temp.append(sm.tsa.arima.ARIMA(data,order=(p,1,q)).fit().bic)
bic_matrix.append(temp)
bic_matrix = pd.DataFrame(bic_matrix) #将其转换成Dataframe 数据结构
print("bic_matrix",bic_matrix)
p,q = bic_matrix.stack().astype(float).idxmin() #先使用stack 展平, 然后使用 idxmin 找出最小值的位置
print(u'BIC 最小的p值 和 q 值:%s,%s' %(p,q)) # BIC 最小的p值 和 q 值:0,1
model = sm.tsa.ARIMA(data, order=(p,1,q)).fit()
model.summary() #生成一份模型报告
predictions_ARIMA_diff = pd.Series(model.fittedvalues, copy=True)
print(predictions_ARIMA_diff)
model.forecast(5) #为未来5天进行预测, 返回预测结果, 标准误差, 和置信区间
for index, column in df.iteritems():
if index==indexz:
continue
index1,index2 =f(column)
length=len(column)
# print("index1 index2:",index1,index2)
# print(column[length-index2-1:length-index1])
print(index)
df[index]=df[index].replace( "null",0)
df[index].astype('float')
df[str(index)+"diff1"]=df[index].diff(1)
df[str(index)+"diff2"]=df[index+"diff1"].diff(1)
# 一阶差分还原
# tmpdata2:原数据
# pred:一阶差分后的预测数据
#df_shift = tmpdata2['ecpm_tomorrow'].shift(1)
#predict = pred.add(df_shift)
# predict = pred + df_shift
# print(index2-index1)
#print(df[["时间(年)",index]][length-index2:length-index1])
adf_test(df[[index]][length-index2:length-index1])
box_pierce_test(df[[index]][length-index2:length-index1])
model_fit(df[[index]][length-index2:length-index1],df,index,length,index1,index2)
## model_fit(data,p,q)
stability_judgment(df[[index]][length-index2:length-index1])
stability_judgment(df[[str(index)+"diff1"]][length-index2:length-index1])
# stability_judgment(df[[str(index)+"diff2"]][length-index2:length-index1])
plt.plot(df[["时间(年)"]][length-index2:length-index1],df[[index]][length-index2:length-index1],label="diff0")
plt.plot(df[["时间(年)"]][length-index2:length-index1],df[[str(index)+"diff1"]][length-index2:length-index1],label="diff1")
# plt.plot(df[["时间(年)"]][length-index2:length-index1],df[[str(index)+"diff2"]][length-index2:length-index1],label="diff2")
# df[["时间(年)",index]][length-index2:length-index1].plot(x=indexz,y=index,figsize=(9,9))
plt.xlabel("时间(年)")
plt.ylabel(index)
plt.legend()
plt.show()
os.system("pause")
运行结果如下:
大家可
大家可以学习一下哈。