使用预测数据进行数据操作答案

【问题标题】：Data manipulation with forecasting data使用预测数据进行数据操作
【发布时间】：2023-03-23 23:25:01
【问题描述】：

我正在尝试对两个商店的销售额进行预测：商店 1 和商店 2。就像使用预测包进行预测的结果一样，我得到了这两个表。第一个表包含有关 MAPE 错误的数据，每个模型分别包含（列值）。下面你可以看到数据和数据的屏幕截图。

Table_1<-structure(list(...1 = c("1", "2", "3", "4", "5", "6", "7", "8", 
                        "9", "10", "11", "12", "13", "14", "15", "16", "17", "18", "19", 
                        "20"), X1 = c("SNAIVE", "HW", "ETS", "ARIMA", "STL", "TBATS", 
                                      "NNETAR", "RWF", "TSLM", "FOURIER", "SNAIVE", "HW", "ETS", "ARIMA", 
                                      "STL", "TBATS", "NNETAR", "RWF", "TSLM", "FOURIER"), X2 = c("Store 1", 
                                                                                                  "Store 1", "Store 1", "Store 1", "Store 1", "Store 1", "Store 1", 
                                                                                                  "Store 1", "Store 1", "Store 1", "Store 2", "Store 2", "Store 2", 
                                                                                                  "Store 2", "Store 2", "Store 2", "Store 2", "Store 2", "Store 2", 
                                                                                                  "Store 2"), value = c(11.2819379803024, 4.90469397146697, 4.90469397146697, 
                                                                                                                        4.64808116952175, 4.92695563666538, 6.11286061911487, 7.66061575087076, 
                                                                                                                        8.95984865369006, 5.07614708345642, 4.57448859126253, 22.7760224588221, 
                                                                                                                        24.0502857269679, 18.9376978459644, 21.6693712888351, 21.6029490199174, 
                                                                                                                        24.692214948761, 26.2680955559159, 30.5302345480261, 22.2367412218357, 
                                                                                                                        22.6100823447494)), row.names = c(NA, -20L), class = c("tbl_df", 
                                                                                                                                                                               "tbl", "data.frame"))

我已经用黄色突出显示了对 Store 1 和 Store 2 的预测中最好的三个模型（MAPE 误差最低）。对于 Store 1 是（ETS、ARIMA 和 Fourier），对于 Store 2（ETS、ARIMA 和 STL）。

第二个表格包含所有这些模型的每个月的预测数据。您可以在下面看到数据以及数据的屏幕截图。

   Table2<-structure(list(Date = structure(c(1575158400, 1577836800, 1580515200, 
                                       1583020800, 1585699200, 1588291200, 1590969600, 1593561600, 1596240000, 
                                       1598918400, 1601510400, 1604188800, 1606780800, 1575158400, 1577836800, 
                                       1580515200, 1583020800, 1585699200, 1588291200, 1590969600, 1593561600, 
                                       1596240000, 1598918400, 1601510400, 1604188800, 1606780800, 1575158400, 
                                       1577836800, 1580515200, 1583020800, 1585699200, 1588291200, 1590969600, 
                                       1593561600, 1596240000, 1598918400, 1601510400, 1604188800, 1606780800, 
                                       1575158400, 1577836800, 1580515200, 1583020800, 1585699200, 1588291200, 
                                       1590969600, 1593561600, 1596240000, 1598918400, 1601510400, 1604188800, 
                                       1606780800, 1575158400, 1577836800, 1580515200, 1583020800, 1585699200, 
                                       1588291200, 1590969600, 1593561600, 1596240000, 1598918400, 1601510400, 
                                       1604188800, 1606780800, 1575158400, 1577836800, 1580515200, 1583020800, 
                                       1585699200, 1588291200, 1590969600, 1593561600, 1596240000, 1598918400, 
                                       1601510400, 1604188800, 1606780800, 1575158400, 1577836800, 1580515200, 
                                       1583020800, 1585699200, 1588291200, 1590969600, 1593561600, 1596240000, 
                                       1598918400, 1601510400, 1604188800, 1606780800, 1575158400, 1577836800, 
                                       1580515200, 1583020800, 1585699200, 1588291200, 1590969600, 1593561600, 
                                       1596240000, 1598918400, 1601510400, 1604188800, 1606780800, 1575158400, 
                                       1577836800, 1580515200, 1583020800, 1585699200, 1588291200, 1590969600, 
                                       1593561600, 1596240000, 1598918400, 1601510400, 1604188800, 1606780800, 
                                       1575158400, 1577836800, 1580515200, 1583020800, 1585699200, 1588291200, 
                                       1590969600, 1593561600, 1596240000, 1598918400, 1601510400, 1604188800, 
                                       1606780800), class = c("POSIXct", "POSIXt"), tzone = "UTC"), 
                    Forecasting_model = c("SNAIVE", "SNAIVE", "SNAIVE", "SNAIVE", 
                                          "SNAIVE", "SNAIVE", "SNAIVE", "SNAIVE", "SNAIVE", "SNAIVE", 
                                          "SNAIVE", "SNAIVE", "SNAIVE", "HW", "HW", "HW", "HW", "HW", 
                                          "HW", "HW", "HW", "HW", "HW", "HW", "HW", "HW", "ETS", "ETS", 
                                          "ETS", "ETS", "ETS", "ETS", "ETS", "ETS", "ETS", "ETS", "ETS", 
                                          "ETS", "ETS", "ARIMA", "ARIMA", "ARIMA", "ARIMA", "ARIMA", 
                                          "ARIMA", "ARIMA", "ARIMA", "ARIMA", "ARIMA", "ARIMA", "ARIMA", 
                                          "ARIMA", "STL", "STL", "STL", "STL", "STL", "STL", "STL", 
                                          "STL", "STL", "STL", "STL", "STL", "STL", "TBATS", "TBATS", 
                                          "TBATS", "TBATS", "TBATS", "TBATS", "TBATS", "TBATS", "TBATS", 
                                          "TBATS", "TBATS", "TBATS", "TBATS", "NNAR", "NNAR", "NNAR", 
                                          "NNAR", "NNAR", "NNAR", "NNAR", "NNAR", "NNAR", "NNAR", "NNAR", 
                                          "NNAR", "NNAR", "RWF", "RWF", "RWF", "RWF", "RWF", "RWF", 
                                          "RWF", "RWF", "RWF", "RWF", "RWF", "RWF", "RWF", "TSLM", 
                                          "TSLM", "TSLM", "TSLM", "TSLM", "TSLM", "TSLM", "TSLM", "TSLM", 
                                          "TSLM", "TSLM", "TSLM", "TSLM", "FOURIER", "FOURIER", "FOURIER", 
                                          "FOURIER", "FOURIER", "FOURIER", "FOURIER", "FOURIER", "FOURIER", 
                                          "FOURIER", "FOURIER", "FOURIER", "FOURIER"), `Store 1` = c(8083, 
                                                                                                     1171, 1328, 1281, 1281, 1118, 1107, 1611, 1116, 1133, 1618, 
                                                                                                     1261, 8083, 8312, 1336, 1261, 1673, 1667, 1223, 1603, 1621, 
                                                                                                     1211, 1633, 1637, 1672, 8138, 8312, 1336, 1261, 1673, 1667, 
                                                                                                     1223, 1603, 1621, 1211, 1633, 1637, 1672, 8138, 8818, 1363, 
                                                                                                     1282, 1671, 1623, 1276, 1283, 1687, 1261, 1632, 1676, 1631, 
                                                                                                     8367, 8827, 1108, 1226, 1681, 1661, 1288, 1616, 1683, 1278, 
                                                                                                     1663, 1678, 1703, 8338, 8371, 1183, 1237, 1738, 1701, 1637, 
                                                                                                     1681, 1721, 1271, 1738, 1663, 1732, 8180, 8076, 1318, 1271, 
                                                                                                     1732, 1883, 1286, 1607, 1336, 1281, 1711, 1873, 1881, 8183, 
                                                                                                     1271, 1283, 1233, 1608, 1618, 1681, 1631, 1611, 1620, 1660, 
                                                                                                     1663, 1673, 1688, 8166, 1317, 1188, 1233, 1273, 1183, 1212, 
                                                                                                     1276, 1178, 1221, 1226, 1283, 8863, 8811, 1118, 1223, 1661, 
                                                                                                     1621, 1260, 1286, 1617, 1213, 1688, 1687, 1660, 8311), `Store 2` = c(1180, 
                                                                                                                                                                          811, 312, 1612, 1387, 878, 812, 883, 362, 768, 800, 760, 
                                                                                                                                                                          1180, 1021, 761, 1002, 1106, 1271, 337, 1113, 373, 833, 1012, 
                                                                                                                                                                          333, 303, 1166, 336, 708, 332, 1312, 1168, 838, 1010, 862, 
                                                                                                                                                                          773, 883, 861, 767, 1000, 1070, 636, 838, 1161, 1183, 887, 
                                                                                                                                                                          1001, 813, 331, 820, 738, 732, 1087, 333, 688, 810, 1311, 
                                                                                                                                                                          1183, 876, 338, 818, 816, 818, 816, 773, 333, 337, 888, 871, 
                                                                                                                                                                          1378, 1100, 1008, 368, 380, 883, 386, 872, 838, 363, 1102, 
                                                                                                                                                                          301, 831, 1133, 1331, 831, 333, 321, 338, 883, 832, 881, 
                                                                                                                                                                          1303, 766, 778, 773, 782, 731, 737, 801, 810, 816, 888, 883, 
                                                                                                                                                                          832, 811, 1820, 1000, 1136, 1270, 1718, 1188, 1873, 1162, 
                                                                                                                                                                          1136, 1130, 1178, 1110, 1371, 380, 703, 306, 1862, 1110, 
                                                                                                                                                                          873, 327, 837, 808, 817, 838, 726, 371)), row.names = c(NA, 
                                                                                                                                                                                                                                  -130L), class = c("tbl_df", "tbl", "data.frame"))

因此，我的意图是根据最低 MAPE 误差自动选择最佳三个模型，例如上面突出显示的模型，并按月计算商店 1 和商店 2 的最佳三个模型的平均值，如下例所示。

我尝试使用此代码，但不知道如何继续。

# Arrange data by MAPE error
Table_1a<-data.frame(Table_1)%>%
  select(X1,X2,value)%>%
  arrange((value),.by_group = TRUE)

# Select three best models
Table_1b <-data.frame(rbind(Table_1a[1:3, 1:3],Table_1a[10:13, 1:3]))%>%
  select(X1,X2)%>%
  group_by(X1,X2)

# Тhis line does not work 

Forecasting_Store_1<-mutate(Table_2,
                            ifelse(Table_1b$X1==Table_2$Forecasting_model,Table_2$Forecasting_model,"")
                         )

那么谁能帮我解决这个问题？

【问题讨论】：

看起来您的标签走在了正确的轨道上。使用Table_1 选择要按商店使用的模型名称，然后使用这些名称从Table2 的长版本中选择行，以按商店和日期进行平均。你试过什么？
我用额外的代码更新了我的问题。我尝试使用包 dplyr。我已经选择了三个最佳模型。下一个陡峭的应该是与表 2 的选择并提取三个最佳模型的平均值。但是我停在这里，不知道如何进行

标签： r dplyr reshape reshape2

【解决方案1】：

这是一个可能的解决方案：

首先，您选择 3 个最佳模型。我更喜欢使用top_n，它与您的解决方案类似，但更简洁。诀窍是粘贴模型和商店以获得唯一键。

model_ok = Table_1 %>% 
  group_by(X2) %>% 
  top_n(-3, value) %>% ungroup %>% 
  transmute(model_ok=paste(X1,X2)) %>% unlist

请注意，在您的示例示例中，第三位并列，因此我的代码为商店 1 选择了 4 个模型而不是 3（您的代码也是如此）。

然后您可以旋转第二个表以将商店存储在行而不是列中，再次粘贴并过滤与接受的键匹配的行。

table3=Table2 %>% 
  pivot_longer(c(`Store 1`,`Store 2`), names_to = "store") %>% 
  mutate(model_store=paste(Forecasting_model, store)) %>% 
  filter(model_store %in% model_ok) %>% 
  select(-model_store)

最后，您可以再次旋转表格以获取模型作为列并计算 3 个模型（或在我的情况下为 4 个）的平均值。如果您只有 2 家商店，您可以使用“Store 2”重复此代码。

table3 %>% 
  filter(store=="Store 1") %>% 
  pivot_wider(names_from = Forecasting_model) %>% 
  mutate(average=rowMeans(select(., -Date, -store)))

编辑：

由于您似乎有多个商店，这里有一个示例，说明如何使用purrr::map 循环访问这些商店。首先，您需要将不同的商店作为命名向量。我使用了Table_1$X2 %>% unique %>% set_names，但您可能想使用更干净的对象。

library(purrrr)
output=Table_1$X2 %>% unique %>% set_names %>%  map(~{
  table3 %>% 
    filter(store==.x) %>% 
    pivot_wider(names_from = Forecasting_model) %>% 
    mutate(average=rowMeans(select(., -Date, -store)))
})
output$`Store 1`
output$`Store 2`

希望对您有所帮助。

【讨论】：

感谢您的代码。我尝试过，乍一看代码工作正常。您的代码选择了四个系列，但这不是问题，因为两个系列具有相同的值。此外，在您的回答中，您提到了循环和咕噜声： :map for loop over stores。实际上我有很多商店可供预测，所以请您添加该选项吗？
@silent_hunter 你去 :-)
感谢您的帮助。这段代码终于可以工作了 :)