Inspect Time Series Cross Validation

Inspect Time Series Cross Validation

suppressPackageStartupMessages(library(tidyverse))
library(targets)
library(DT)

knitr::opts_knit$set(root.dir = "../../")

Strategy

tar_read(p_tscv)

Predictions

Test set predictions (blue bars) of each iteration.

df_cv_preds_and_coefs <- tar_read(df_cv_preds_and_coefs)

df_cv_preds_and_coefs

## # A tibble: 45 × 6
##    year_start_act category_id index_past  index_next_year formula_str      pred 
##             <dbl> <chr>       <list>      <list>          <chr>            <lis>
##  1           2011 cnsn        <int [128]> <int [130]>     has_finding ~ n… <tib…
##  2           2012 cnsn        <int [258]> <int [105]>     has_finding ~ n… <tib…
##  3           2013 cnsn        <int [363]> <int [85]>      has_finding ~ n… <tib…
##  4           2014 cnsn        <int [448]> <int [98]>      has_finding ~ n… <tib…
##  5           2015 cnsn        <int [546]> <int [94]>      has_finding ~ n… <tib…
##  6           2016 cnsn        <int [640]> <int [83]>      has_finding ~ n… <tib…
##  7           2017 cnsn        <int [723]> <int [46]>      has_finding ~ n… <tib…
##  8           2018 cnsn        <int [769]> <int [39]>      has_finding ~ n… <tib…
##  9           2019 cnsn        <int [808]> <int [3]>       has_finding ~ n… <tib…
## 10           2015 dtin        <int [546]> <int [94]>      has_finding ~ n… <tib…
## # … with 35 more rows

pred <- df_cv_preds_and_coefs$pred[[1]]

pred

## # A tibble: 1 × 3
##   pred_train         pred_valid         coefs            
##   <list>             <list>             <list>           
## 1 <tibble [128 × 3]> <tibble [130 × 3]> <tibble [10 × 8]>

df_cv_preds_and_coefs %>%
  select(year_start_act, category_id, pred) %>%
  mutate(pred_valid = map(pred, "pred_valid")) %>%
  select(- pred) %>%
  unnest(pred_valid) %>%
  unnest(pred_valid)

## # A tibble: 3,089 × 5
##    year_start_act category_id activity_id_new pred_yes   obs
##             <dbl> <chr>       <chr>              <dbl> <dbl>
##  1           2011 cnsn        00203              0.418     0
##  2           2011 cnsn        00318              0.277     0
##  3           2011 cnsn        00319              0.558     0
##  4           2011 cnsn        00329              0.304     1
##  5           2011 cnsn        00348              0.471     0
##  6           2011 cnsn        00355              0.471     0
##  7           2011 cnsn        00362              0.471     0
##  8           2011 cnsn        00368              0.320     1
##  9           2011 cnsn        00369              0.550     0
## 10           2011 cnsn        00370              0.320     1
## # … with 3,079 more rows

Coefficients

Coefficients and coefficient statistics of each iteration.

df_cv_preds_and_coefs %>%
  select(year_start_act, category_id, pred) %>%
  mutate(coefs = map(pred, "coefs")) %>%
  select(- pred) %>%
  unnest(coefs) %>%
  unnest(coefs) %>%
  DT::datatable()