pyhon 進階分析 - 4.1 時間序列 | M5 資料簡介 ( M5 Time Series in python )

[@play_data]

# 完整程式
import sys
sys.path.append(r'D:\Users\Pu_chang\Desktop\資料分析\UsefulML')
from PreProcessing import TS_PreProcess, External_Variable
import datetime
import itertools # grid_search
import matplotlib.pyplot as plt
import numpy as np
import pandas as pd
from prophet import Prophet
from prophet.diagnostics import performance_metrics, cross_validation
import warnings
warnings.filterwarnings('ignore')
'每個產品於每個商店銷量，時間 = d_1 ~ d_1913'
path = 'D:/Users/Pu_chang/Desktop/資料分析/3. 推論分析/時間序列/data/M5/'
validation = pd.read_csv(path + 'sales_train_validation.csv', encoding='utf-8',header=0)
calender = pd.read_csv(path + 'calendar.csv', encoding='utf-8',header=0)
evaluation = pd.read_csv(path + 'sales_train_evaluation.csv', encoding='utf-8',header=0)
greater_than_zero = (validation.iloc[:,6:] > 0).sum(axis=1).reset_index()
'易 : 全 > 0的跑，5859，2011-01-29 ~ 2016-04-24'
def split_train_test(item, df1, df2) :
train = df1.loc[item].reset_index().iloc[6:,:]
train.columns = ['d', 'y']
train['y'] = train['y'].astype(int)
train = pd.merge(train, calender, on="d",how='inner')
train = train.sort_values(by = 'date', ascending=True)

test = df2.loc[item].reset_index().iloc[6:,:]
test.columns = ['d', 'y']
test['y'] = test['y'].astype(int)
test = pd.merge(test, calender, on="d",how='inner')
test['d'] = test['d'].replace(to_replace ='d_', value = '', regex = True).astype(int)
test = test.loc[test['d']>=1914]
test = test.sort_values(by = 'date', ascending=True)
return train, test
train, test = split_train_test(5859, validation, evaluation)