更新R包后parsnip模型调用modeltime_calibrate函数报错

开发的一组时间序列预测模型，更新包括modeltime、timetk在内的多个R包前代码可完全正常运行；更新包后代码运行异常，经排查错误出现在校准步骤执行阶段，涉及ranger与xgboost两类模型。
核心报错提示：从新数据获取预测变量时出错，tk_get_timeseries_signature.default无适用于NULL类的方法，仅ranger、xgboost校准失败，prophet模型运行正常。

包更新前可正常运行代码的会话环境信息如下：

library(forecast)
library(tidyverse)
library(lubridate)
library(quantdates)
library(tidymodels)
library(timetk)
library(modeltime)
library(modeltime.ensemble)

> sessionInfo()
R version 4.1.3 (2022-03-10)
Platform: x86_64-w64-mingw32/x64 (64-bit)
Running under: Windows 10 x64 (build 19044)

Matrix products: default

attached base packages:
[1] stats     graphics  grDevices utils     datasets  methods   base     

other attached packages:
 [1] kernlab_0.9-30           prophet_1.0              rlang_1.0.2              Rcpp_1.0.8.3             xgboost_1.6.0.1          ranger_0.13.1           
 [7] modeltime.ensemble_1.0.0 modeltime.resample_0.2.0 modeltime_1.2.0          timetk_2.8.0             yardstick_0.0.9          workflowsets_0.2.1      
[13] workflows_0.2.6          tune_0.2.0               rsample_0.1.1            recipes_0.2.0            parsnip_0.2.1            modeldata_0.1.1         
[19] infer_1.0.0              dials_0.1.1              scales_1.2.0             broom_0.8.0              tidymodels_0.2.0         quantdates_1.0          
[25] lubridate_1.8.0          forcats_0.5.1            stringr_1.4.0            dplyr_1.0.9              purrr_0.3.4              readr_2.1.2             
[31] tidyr_1.2.0              tibble_3.1.7             ggplot2_3.3.6            tidyverse_1.3.1          forecast_8.16           

loaded via a namespace (and not attached):
  [1] readxl_1.4.0         backports_1.4.1      plyr_1.8.7           splines_4.1.3        listenv_0.8.0        inline_0.3.19        digest_0.6.29       
  [8] foreach_1.5.2        fansi_1.0.3          magrittr_2.0.3       tzdb_0.3.0           Metrics_0.1.4        globals_0.15.0       modelr_0.1.8        
 [15] gower_1.0.0          RcppParallel_5.1.5   matrixStats_0.62.0   xts_0.12.1           hardhat_0.2.0        tseries_0.10-51      prettyunits_1.1.1   
 [22] colorspace_2.0-3     rvest_1.0.2          warp_0.2.0           haven_2.5.0          callr_3.7.0          crayon_1.5.1         jsonlite_1.8.0      
 [29] progressr_0.10.0     survival_3.3-1       zoo_1.8-10           iterators_1.0.14     glue_1.6.2           gtable_0.3.0         ipred_0.9-12        
 [36] pkgbuild_1.3.1       rstan_2.21.5         future.apply_1.9.0   quantmod_0.4.20      DBI_1.1.2            GPfit_1.0-8          stats4_4.1.3        
 [43] lava_1.6.10          StanHeaders_2.21.0-7 prodlim_2019.11.13   httr_1.4.3           ellipsis_0.3.2       pkgconfig_2.0.3      loo_2.5.1           
 [50] nnet_7.3-17          dbplyr_2.1.1         utf8_1.2.2           janitor_2.1.0        tidyselect_1.1.2     DiceDesign_1.9       munsell_0.5.0       
 [57] cellranger_1.1.0     tools_4.1.3          cli_3.2.0            generics_0.1.2       processx_3.5.3       fs_1.5.2             future_1.25.0       
 [64] nlme_3.1-155         tictoc_1.0.1         xml2_1.3.3           compiler_4.1.3       rstudioapi_0.13      curl_4.3.2           slider_0.2.2        
 [71] reprex_2.0.1         lhs_1.1.5            stringi_1.7.6        ps_1.7.0             lattice_0.20-45      Matrix_1.4-1         urca_1.3-0          
 [78] vctrs_0.4.1          pillar_1.7.0         lifecycle_1.0.1      furrr_0.3.0          lmtest_0.9-40        data.table_1.14.2    R6_2.5.1            
 [85] gridExtra_2.3        parallelly_1.31.1    codetools_0.2-18     MASS_7.3-55          assertthat_0.2.1     withr_2.5.0          fracdiff_1.5-1      
 [92] parallel_4.1.3       hms_1.1.1            quadprog_1.5-8       grid_4.1.3           rpart_4.1.16         timeDate_3043.102    class_7.3-20        
 [99] snakecase_0.11.0     TTR_0.24.3           pROC_1.18.0  

更新包后运行以下代码触发报错：

wflw_mod_rf <- workflow() %>%
    add_model(
      spec = rand_forest(
        mode = "regression"
      ) %>%
        set_engine("ranger")
    ) %>%
    add_recipe(recipe_spec %>%
                 update_role(dtemonth, new_role = "indicator")) %>%
    fit(training(splits))
 
  wflw_mod_xgboost <- workflow() %>%
    add_model(
      spec = boost_tree(
        mode = "regression"
      ) %>%
        set_engine("xgboost")
    ) %>%
    add_recipe(recipe_spec %>%
                 update_role(dtemonth, new_role = "indicator")) %>%
    fit(training(splits))
 
  wflw_mod_prophet <- workflow() %>%
    add_model(
      spec = prophet_reg(
        seasonality_daily  = FALSE,
        seasonality_weekly = FALSE,
        seasonality_yearly = TRUE
      ) %>%
        set_engine("prophet")
    ) %>%
    add_recipe(recipe_spec) %>%
    fit(training(splits))

submodels_all_tbl <- modeltime_table(
    wflw_mod_rf,
    wflw_mod_xgboost,
    wflw_mod_prophet
  )
  
> submodels_all_tbl %>% modeltime_calibrate(testing(splits), quiet = F)
Error: Problem occurred getting predictors from new data. Error in tk_get_timeseries_signature.default(.): No method for class NULL.

Error: Problem occurred getting predictors from new data. Error in tk_get_timeseries_signature.default(.): No method for class NULL.



── Model Calibration Failure Report ────────────────────────
# A tibble: 2 × 6
  .model_id .model     .model_desc .type .calibration_data fail_check
      <int> <list>     <chr>       <chr> <list>            <lgl>    
1         1 <workflow> RANGER      NA    <lgl [1]>         TRUE      
2         2 <workflow> XGBOOST     NA    <lgl [1]>         TRUE      
The following models had errors:
- Model 1: Failed Calibration.
- Model 2: Failed Calibration.

Potential Solution: Check the Error/Warning Messages for clues as to why your model(s) failed calibration.
── End Model Calibration Failure Report ────────────────────

该报错是modeltime、timetk版本迭代后的逻辑变更导致：

• 旧版本（modeltime 1.2.0、timetk 2.8.0）不会强制要求ranger、xgboost这类纯机器学习模型的工作流中保留显式标记的时间索引列
• 新版本在校准机器学习类时间序列模型时，会自动提取数据中的时间索引列做时间特征合法性校验，代码中将时间列dtemonth的角色更新为indicator，导致框架无法识别到合法的时间索引，传入tk_get_timeseries_signature的对象为NULL，最终触发报错
• prophet模型自带内置的时间索引解析逻辑，不需要从recipe中识别时间列，因此可以正常运行

• 方案1（推荐）：将ranger、xgboost工作流中dtemonth列的角色显式指定为"index"，该角色下的列不会被当作特征输入模型，不会影响原有模型逻辑。修正后的代码示例：

wflw_mod_rf <- workflow() %>%
  add_model(
    spec = rand_forest(mode = "regression") %>%
      set_engine("ranger")
  ) %>%
  add_recipe(recipe_spec %>%
               update_role(dtemonth, new_role = "index")) %>%
  fit(training(splits))

xgboost工作流做相同修改即可。

• 方案2：如果需要将dtemonth衍生的时间特征作为模型输入，可在recipe中用step_timeseries_signature()提前生成需要的时间维度特征，再将原始dtemonth列的角色设为"index"即可。
• 临时兼容方案：如果不想修改现有代码，可将modeltime回退到1.2.0版本、timetk回退到2.8.0版本，即可恢复原有运行逻辑，但不推荐长期使用旧版本。

内容的提问来源于stack exchange，提问作者Jean Paul PG