如何处理modeltime预测中melt及unmelt操作导致的变量丢失问题

• 分组逻辑缺失：当前将240个变量融化后的长表直接作为输入训练单一Prophet模型，Modeltime的单序列模型默认不会自动区分variable分组，相当于把所有变量的数值合并为单条时间序列训练，预测结果自然无法覆盖全部240个变量。
• 时间拆分逻辑错误：直接对全量长表执行initial_time_split是按行顺序切割前80%为训练集，而非按每个变量单独按时间维度拆分训练/测试集，导致大量变量的2020年区间数据根本没有进入测试集，预测结果不存在这些变量的记录，宽表转换时就会丢失。
• 预测值列取值错误：modeltime_forecast输出的预测值存储在.value字段，你之前提取的value是原始实际观测值，取值错误也会导致宽表转换异常。

采用分组嵌套的方式对每个变量单独训练模型、预测，保证所有变量都有对应的预测结果，修正后代码如下：

suppressPackageStartupMessages(library(tidyverse))
suppressPackageStartupMessages(library(lubridate))
suppressPackageStartupMessages(library(tidymodels))
suppressPackageStartupMessages(library(modeltime))

## 生成模拟数据
dates <- ymd("2016-01-01") + months(0:59)
fake_values <- 
  c(661,678,1094,1987,3310,2105,1452,983,1107,805,675,684,436,514,668,206,19,23,365,456,1174,1760,735,366,
    510,580,939,1127,2397,1514,1370,832,765,661,497,328,566,631,983,1876,2784,2928,2543,1508,1175,8,1733,
    862,779,1112,1446,2407,3917,2681,2397,1246,1125,1223,1234,1239,
    661,678,1094,1987,3310,2105,1452,983,1107,805,675,684,436,514,668,206,19,23,365,456,1174,1760,735,366,
    510,580,939,1127,2397,1514,1370,832,765,661,497,328,566,631,983,1876,2784,2928,2543,1508,1175,8,1733,
    862,779,1112,1446,2407,3917,2681,2397,1246,1125,1223,1234,1239,
    661,678,1094,1987,3310,2105,1452,983,1107,805,675,684,436,514,668,206,19,23,365,456,1174,1760,735,366,
    510,580,939,1127,2397,1514,1370,832,765,661,497,328,566,631,983,1876,2784,2928,2543,1508,1175,8,1733,
    862,779,1112,1446,2407,3917,2681,2397,1246,1125,1223,1234,1239,
    661,678,1094,1987,3310,2105,1452,983,1107,805,675,684,436,514,668,206,19,23,365,456,1174,1760,735,366,
    510,580,939,1127,2397,1514,1370,832,765,661,497,328,566,631,983,1876,2784,2928,2543,1508,1175,8,1733,
    862,779,1112,1446,2407,3917,2681,2397,1246,1125,1223,1234,1239)

replicate <- rep(1,60) %*% t.default(fake_values)
replicate <- as.data.frame(replicate)

df <- bind_cols(replicate, dates) %>%
  rename(dates = ...241)

## 转换为长表
data <- reshape2::melt(df, id.var='dates')

## 按变量分组嵌套，每个变量对应独立时间序列
data_nested <- data %>%
  group_by(variable) %>%
  nest()

## 定义单序列预测函数
forecast_single <- function(df) {
  # 分组内按时间拆分训练测试集
  split_obj <- initial_time_split(df, prop = 0.8)
  # 训练Prophet模型
  model_fit <- prophet_reg() %>%
    set_engine("prophet") %>%
    fit(value ~ dates, data = training(split_obj))
  # 生成预测结果
  modeltime_table(model_fit) %>%
    modeltime_forecast(
      new_data = testing(split_obj),
      actual_data = df,
      keep_data = TRUE
    ) %>%
    filter(str_detect(.key, "prediction"))
}

## 批量对所有变量执行预测
fc_all <- data_nested %>%
  mutate(forecast_res = map(data, forecast_single)) %>%
  unnest(forecast_res)

## 筛选2020年数据并转换为宽表
fc_wide_prophet <- fc_all %>%
  filter(.index >= ymd("2020-01-01"), .index <= ymd("2020-12-01")) %>%
  select(.index, variable, .value) %>%
  pivot_wider(names_from = variable, values_from = .value)

内容的提问来源于stack exchange，提问作者LGe