data processing.zip_electricityzoo_gentle61j_kaggle_sales-predic

# -*- coding:utf-8 –*- import pandas as pd import datetime import numpy as np from sklearn.model_selection import train_test_split import matplotlib.pyplot as plt from sklearn import ensemble from sklearn.model_selection import cross_val_score from sklearn.preprocessing import Imputer def dataProcessing(): train = pd.read_csv(r'E:\pycharm files\sales prediction\data\train.csv') feature = pd.read_csv(r'E:\pycharm files\sales prediction\data\features.csv') # 统一Date格式 feature['Date'] = [datetime.datetime.strptime(x, '%Y/%m/%d').date() for x in feature['Date']] train['Date'] = [datetime.datetime.strptime(y, '%Y-%m-%d').date() for y in train['Date']] # 将train和feature整合为一张表 traindataNew = pd.merge(train, feature, how='left', on=['Date', 'Store', 'IsHoliday']) # 将IsHoliday的布尔值化为0，1 traindataNew['IsHoliday'] = [conVertbool(x) for x in traindataNew['IsHoliday']] # 将日期化为回归可接受的数字 weekNm = [(x - traindataNew['Date'][0]) for x in list(traindataNew['Date'])] traindataNew['Date'] = [(float(np.timedelta64(x, 'D').astype(int) / 7 % 52)) for x in weekNm] # 归一化 traindataNew['Temperature'] = regularization(list(traindataNew['Temperature'])) traindataNew['Fuel_Price'] = regularization(list(traindataNew['Fuel_Price'])) traindataNew['CPI'] = regularization(list(traindataNew['CPI'])) traindataNew['Unemployment'] = regularization(list(traindataNew['Unemployment'])) # 去除5个markdown列,去除特征'CPI'和'Unemployment'，因为测试集中缺失值 traindataNew = traindataNew.iloc[:, [0, 1, 2, 3, 4, 5, 6]] # 将45个store转化为45个特征 Store_Dummies = pd.get_dummies(traindataNew.Store, prefix='Store').iloc[:, 0:] traindataNewchange = pd.concat([traindataNew, Store_Dummies], axis=1) # 将99个部门转化为特征 Dept_Dummies = pd.get_dummies(traindataNewchange.Dept, prefix='Dept').iloc[:, 0:] traindataNewchange2 = pd.concat([traindataNewchange, Dept_Dummies], axis=1) dataNew = traindataNewchange2.to_csv(r'E:\pycharm files\sales prediction\data\traindataNew.csv') return dataNew # 将布尔值转化为0，1 def conVertbool(x): if x: return 1 else: return 0 #归一化 def regularization(List): Max = max(List) Min = min(List) List = [(x - Min) / (Max - Min) for x in List] return List # 训练模型 def train_modle(): traindataNew = pd.read_csv(r'E:\pycharm files\sales prediction\data\traindataNew.csv') # 从样本中随机选取训练样本和测试样本 train_sample, test_sample = train_test_split(traindataNew, test_size=0.2, random_state=42) xtrain = train_sample.iloc[:, ([3] + range(5, 136))] ytrain = train_sample.Weekly_Sales xtest = train_sample.iloc[:, ([3] + range(5, 136))] ytest = train_sample.Weekly_Sales # 调用随机森林 wmLinear = ensemble.RandomForestRegressor(n_estimators=25, oob_score=True) wmLinear.fit(xtrain, ytrain) score = wmLinear.score(xtest, ytest) print (score) ypredict = wmLinear.predict(xtest) plt.plot(ytest, ypredict, 'ro') plt.plot(ytest, ytest, 'b-') plt.show() return wmLinear def testdata_processing(): train = pd.read_csv(r'E:\pycharm files\sales prediction\data\train.csv') testdata = pd.read_csv(r'E:\pycharm files\sales prediction\data\test.csv') feature = pd.read_csv(r'E:\pycharm files\sales prediction\data\features.csv') train['Date'] = [datetime.datetime.strptime(y, '%Y-%m-%d').date() for y in train['Date']] feature['Date'] = [datetime.datetime.strptime(x, '%Y/%m/%d').date() for x in feature['Date']] testdata['Date'] = [datetime.datetime.strptime(y, '%Y-%m-%d').date() for y in testdata['Date']] testdataNew = pd.merge(testdata, feature, how='left', on=['Date', 'Store', 'IsHoliday']) testdataNew['IsHoliday'] = [conVertbool(x) for x in testdataNew['IsHoliday']] weekNm = [(x - train['Date'][0]) for x in list(testdataNew['Date'])] testdataNew['Date'] = [(float(np.timedelta64(x, 'D').astype(int) / 7 % 52)) for x in weekNm] #缺失值填充 # testdataNew['CPI'] = testdataNew['CPI'].fillna(method='pad') # testdataNew['Unemployment'] = testdataNew['Unemployment'].fillna(method='pad') testdataNew['Temperature'] = regularization(list(testdataNew['Temperature'])) testdataNew['Fuel_Price'] = regularization(list(testdataNew['Fuel_Price'])) testdataNew['CPI'] = regularization(list(testdataNew['CPI'])) testdataNew['Unemployment'] = regularization(list(testdataNew['Unemployment'])) testdataNew = testdataNew.iloc[:, [0, 1, 2, 3, 4, 5]] Store_Dummies = pd.get_dummies(testdataNew.Store, prefix='Store').iloc[:, 0:] testdataNewchange = pd.concat([testdataNew, Store_Dummies], axis=1) # 将99个部门转化为特征 Dept_Dummies = pd.get_dummies(testdataNewchange.Dept, prefix='Dept').iloc[:, 0:] testdataNewchange2 = pd.concat([testdataNewchange, Dept_Dummies], axis=1) dataNew = testdataNewchange2.to_csv(r'E:\pycharm files\sales prediction\data\testdataNew.csv') return dataNew # K折交叉验证 # dataProcessing() # traindataNew = pd.read_csv(r'E:\pycharm files\sales prediction\data\traindataNew.csv') # xtrain = traindataNew.iloc[:, ([3] + range(5, 136))] # ytrain = traindataNew.Weekly_Sales # wmLinear = ensemble.RandomForestRegressor(n_estimators=25) # score=cross_val_score(wmLinear,xtrain,ytrain,cv=2) # print score # print sum(score)/10 #实际预测 dataProcessing() testdata_processing() testdataNew = pd.read_csv(r'E:\pycharm files\sales prediction\data\testdataNew.csv') xtestt = testdataNew.iloc[:, range(3,133)] traindataNew = pd.read_csv(r'E:\pycharm files\sales prediction\data\traindataNew.csv') xtrain = traindataNew.iloc[:, ([3] + range(5, 134))] ytrain = traindataNew.Weekly_Sales wmLinear = ensemble.RandomForestRegressor(n_estimators=150,oob_score=True,n_jobs=-1) wmLinear.fit(xtrain, ytrain) print (wmLinear.oob_score_) salespredict = wmLinear.predict(xtestt) result = pd.read_csv(r'E:\pycharm files\sales prediction\data\sampleSubmission.csv') result['Weekly_Sales'] = [x for x in salespredict] result.to_csv(r'E:\pycharm files\sales prediction\data\result.csv') # # 袋外分数 # dataProcessing() # traindataNew = pd.read_csv(r'E:\pycharm files\sales prediction\data\traindataNew.csv') # xtrain = traindataNew.iloc[:, ([3] + range(5, 136))] # ytrain = traindataNew.Weekly_Sales # wmLinear = ensemble.RandomForestRegressor(n_estimators=100, oob_score=True) # wmLinear.fit(xtrain, ytrain) # print wmLinear.oob_score_ # #画图 # train = pd.read_csv(r'E:\pycharm files\sales prediction\data\train.csv') # train['Date'] = [datetime.datetime.strptime(y, '%Y-%m-%d').date() for y in train['Date']] # plt.plot(train.iloc[range(0,145),[2]],train.iloc[range(0,145),[3]],'b-') # plt.show() # plt.plot(train.iloc[range(0,52),[2]],train.iloc[range(0,52),[3]],'b-') # plt.show() # plt.plot(train.iloc[range(52,104),[2]],train.iloc[range(52,104),[3]],'b-') # plt.show()