使用 R 进行分层预测

Question

我正在使用寓言包来预测分层时间序列和所有节点的深度不相等。用例是预测国家 -> 州 -> 地区级别的联系人。汇总时预测值必须加起来达到国家级别（较低级别的预测等于较高级别的预测）。
https://robjhyndman.com/papers/Foresight-hts-final.pdf
下面给出的是我在预测测试数据时尝试过的代码。

library(fable)
library(tsibble)
library(tsibbledata)
library(lubridate)
library(dplyr)

# selecting train data
train_df <- tourism %>%
  filter(year(Quarter) <= 2014 & Region %in% c("MacDonnell", "Melbourne"))

# selecting test data
test_df <- tourism %>%
  filter(year(Quarter) > 2014 & Region %in% c("MacDonnell", "Melbourne"))

# fitting ets model with reconcilliation
ets_fit <- train_df %>%
  aggregate_key(Purpose * (State / Region), Trips = sum(Trips)) %>%
  model(ets=ETS(Trips)) %>%
  reconcile(ets_adjusted = min_trace(ets))
# forecasting on test data
fcasts_test <- forecast(ets_fit, test_df)

获取错误为

Error: Provided data contains a different key structure to the models.
Run `rlang::last_error()` to see where the error occurred.

我该如何解决这个问题？

Answer 1

您在拟合模型之前使用 aggregate_key() 更改了键结构，因此预测键结构与测试集不匹配。您需要使用 aggregate_key().

在之后创建测试集
但是，您不能在创建聚合后按其中一个键进行过滤，因为那样聚合信息是不完整的。

这是一个可以满足您要求的示例。

library(fable) library(tsibble) library(tsibbledata) library(lubridate) library(dplyr) # Aggregate data as required agg_tourism <- tourism %>% filter(Region %in% c("MacDonnell", "Melbourne")) %>% aggregate_key(Purpose * (State / Region), Trips = sum(Trips)) # Select training data train_df <- agg_tourism %>% filter(year(Quarter) <= 2014) # Select test data test_df <- agg_tourism %>% filter(year(Quarter) > 2014) # Fit ets model with reconcilliation ets_fit <- train_df %>% model(ets = ETS(Trips)) %>% reconcile(ets_adjusted = min_trace(ets)) # forecasting on test data fcasts_test <- forecast(ets_fit, test_df) fcasts_test #> # A fable: 600 x 7 [1Q] #> # Key: Purpose, State, Region, .model [50] #> Purpose State Region .model Quarter Trips .mean #> <chr*> <chr*> <chr*> <chr> <qtr> <dist> <dbl> #> 1 Business Northern Territory MacDonnell ets 2015 Q1 N(5.1, 21) 5.12 #> 2 Business Northern Territory MacDonnell ets 2015 Q2 N(5.1, 21) 5.12 #> 3 Business Northern Territory MacDonnell ets 2015 Q3 N(5.1, 21) 5.12 #> 4 Business Northern Territory MacDonnell ets 2015 Q4 N(5.1, 21) 5.12 #> 5 Business Northern Territory MacDonnell ets 2016 Q1 N(5.1, 21) 5.12 #> 6 Business Northern Territory MacDonnell ets 2016 Q2 N(5.1, 21) 5.12 #> 7 Business Northern Territory MacDonnell ets 2016 Q3 N(5.1, 21) 5.12 #> 8 Business Northern Territory MacDonnell ets 2016 Q4 N(5.1, 21) 5.12 #> 9 Business Northern Territory MacDonnell ets 2017 Q1 N(5.1, 21) 5.12 #> 10 Business Northern Territory MacDonnell ets 2017 Q2 N(5.1, 21) 5.12 #> # … with 590 more rows fcasts_test %>% filter(Region == "Melbourne", Purpose == "Visiting") %>% autoplot(agg_tourism)

^{由 reprex package (v0.3.0)}
于 2020-12-26 创建

使用 R 进行分层预测

Hierarchical Forecasting using R

r

time-series

hierarchical

forecasting

fable-r