How to export the residuals of a random forest regression while using spatialsample?

[nikos_geo]

I am trying to replicate this tutorial but instead of a linear regression I am using random forest (RF). I can make the prediction but now I would like to compute and extract the residuals of the regression (i.e., observed - predicted values). Then, I would like to cbind the residuals with the coordinates of the data.frame, like so:

resids_df <- cbind(original_df[, 1:2], rf_resids) # where the original_df[, 1:2] contains the coordinates and the rf_resids are the residuals of the RF regression

The issue is that when after making the predictions, the output has 54 more values than my original data set (3792 vs 3738). This causes an issue because I can't cbind the residuals due to the difference in number of rows.

How can I resolve this issue and get exactly the same number of residuals (observations) as my original data set?

P.S. my data set does not contain NA values.

In the example below, I used asubset of my dataset but again, you can see that the there is 1 more value in the predictions compared to the number of rows in the data set.

library(tidymodels)
library(spatialsample)
library(sf)

wd <- "path/"

proj_ref_sys <- "EPSG:7760"

drought <- read.csv(paste0(wd, "block.data.csv"))
nrow(drought)
# [1] 60 !!!!!!!!!!!!!

drought_sf <- st_as_sf(drought, coords = c("x", "y"),  crs = proj_ref_sys)

set.seed(123)
folds <- spatial_block_cv(drought_sf, v = 3)

drought_res <-
  workflow(ntl ~ pop + agbh + nir, 
           rand_forest(mode = "regression", mtry = 2, trees = 100) %>%
             set_engine("randomForest")) %>%
  fit_resamples(folds, control = control_resamples(save_pred = TRUE))

drought_res

collect_predictions(drought_res)

# A tibble: **61** × 5 !!!!!!!!!!!!!!!!!!!
   id    .pred  .row   ntl .config             
   <chr> <dbl> <int> <dbl> <chr>               
 1 Fold1 28.7     18  29.2 Preprocessor1_Model1
 2 Fold1 27.9     19  32.8 Preprocessor1_Model1
 3 Fold1 17.2     20  29.6 Preprocessor1_Model1
 4 Fold1 19.6     21  28.6 Preprocessor1_Model1
 5 Fold1 34.3     22  36.5 Preprocessor1_Model1
 6 Fold1 48.7     28  34.8 Preprocessor1_Model1
 7 Fold1 45.9     29  32.2 Preprocessor1_Model1
 8 Fold1 40.1     30  28.3 Preprocessor1_Model1
 9 Fold1 14.6     31  22.5 Preprocessor1_Model1
10 Fold1  9.96    32  17.1 Preprocessor1_Model1
# ℹ 51 more rows
# ℹ Use `print(n = ...)` to see more rows

The data.frame I'm using:

structure(list(x = c(995494.2549, 995924.2549, 996354.2549, 996784.2549, 
997214.2549, 997644.2549, 998074.2549, 998504.2549, 998934.2549, 
999364.2549, 999794.2549, 1000224.2549, 1000654.2549, 1001084.2549, 
1001514.2549, 1001944.2549, 1002374.2549, 1002804.2549, 1003234.2549, 
1003664.2549, 1004094.2549, 1004524.2549, 1004954.2549, 1005384.2549, 
1005814.2549, 1006244.2549, 1006674.2549, 1007104.2549, 1007534.2549, 
1007964.2549, 1008394.2549, 1008824.2549, 1009254.2549, 1009684.2549, 
1010114.2549, 1010544.2549, 1010974.2549, 1011404.2549, 1011834.2549, 
1012264.2549, 1012694.2549, 1013124.2549, 1013554.2549, 1013984.2549, 
1014414.2549, 1014844.2549, 1015274.2549, 1015704.2549, 1016134.2549, 
1016564.2549, 1016994.2549, 1017424.2549, 1017854.2549, 1018284.2549, 
1018714.2549, 995494.2549, 995924.2549, 996354.2549, 996784.2549, 
997214.2549), y = c(1019851.5842, 1019851.5842, 1019851.5842, 
1019851.5842, 1019851.5842, 1019851.5842, 1019851.5842, 1019851.5842, 
1019851.5842, 1019851.5842, 1019851.5842, 1019851.5842, 1019851.5842, 
1019851.5842, 1019851.5842, 1019851.5842, 1019851.5842, 1019851.5842, 
1019851.5842, 1019851.5842, 1019851.5842, 1019851.5842, 1019851.5842, 
1019851.5842, 1019851.5842, 1019851.5842, 1019851.5842, 1019851.5842, 
1019851.5842, 1019851.5842, 1019851.5842, 1019851.5842, 1019851.5842, 
1019851.5842, 1019851.5842, 1019851.5842, 1019851.5842, 1019851.5842, 
1019851.5842, 1019851.5842, 1019851.5842, 1019851.5842, 1019851.5842, 
1019851.5842, 1019851.5842, 1019851.5842, 1019851.5842, 1019851.5842, 
1019851.5842, 1019851.5842, 1019851.5842, 1019851.5842, 1019851.5842, 
1019851.5842, 1019851.5842, 1019421.5842, 1019421.5842, 1019421.5842, 
1019421.5842, 1019421.5842), ntl = c(9.14866638183594, 15.3856477737427, 
16.3302040100098, 12.454291343689, 10.4823837280273, 11.394606590271, 
8.1963529586792, 4.50725030899048, 3.95374751091003, 5.73203563690186, 
14.3955335617065, 17.0745468139648, 14.2944135665894, 10.333722114563, 
9.80743503570557, 12.5352020263672, 19.8813304901123, 29.2410221099854, 
32.8321876525879, 29.575023651123, 28.5894374847412, 36.4911346435547, 
49.4252128601074, 61.3118171691895, 58.6104736328125, 43.0437355041504, 
28.096061706543, 34.8003845214844, 32.1936340332031, 28.3407783508301, 
22.5178966522217, 17.0638084411621, 20.7549228668213, 18.3547439575195, 
10.2983675003052, 7.3524694442749, 7.17788362503052, 7.06999540328979, 
8.03957176208496, 12.6783542633057, 18.7537479400635, 26.1656856536865, 
36.539493560791, 41.0569839477539, 25.5366401672363, 15.7820110321045, 
9.87918758392334, 7.65169858932495, 6.96318626403809, 8.69833087921143, 
12.1393032073975, 15.151198387146, 14.5944147109985, 9.46016979217529, 
4.53868055343628, 12.8388118743896, 21.1265335083008, 19.3046970367432, 
10.5719947814941, 8.08844661712646), pop = c(31.2753772735596, 
55.8289375305176, 56.4003105163574, 33.795223236084, 31.0511913299561, 
30.5730743408203, 13.667106628418, 7.08161020278931, 6.89333772659302, 
13.9001550674438, 35.5272178649902, 42.4625587463379, 32.9688529968262, 
21.4302787780762, 12.6151924133301, 17.4939270019531, 38.1474113464355, 
60.8120536804199, 65.3665008544922, 53.8765907287598, 46.2705993652344, 
61.42333984375, 70.8307113647461, 53.3152236938477, 31.4083557128906, 
24.9810562133789, 38.3716621398926, 56.114860534668, 67.1656036376953, 
60.8404235839844, 33.7796592712402, 29.8311328887939, 44.3309173583984, 
31.9606342315674, 16.7053775787354, 10.1427822113037, 11.4020376205444, 
10.7794933319092, 18.2773151397705, 34.2912216186523, 50.6655197143555, 
52.1081962585449, 53.0502471923828, 59.4989013671875, 48.5897750854492, 
41.188159942627, 27.0699615478516, 11.5318984985352, 9.09538650512695, 
14.2379903793335, 24.8153190612793, 29.3468627929688, 30.5861835479736, 
15.3130531311035, 9.47307205200195, 37.2332077026367, 94.2268676757812, 
73.2485733032227, 26.8748569488525, 26.8519401550293), agbh = c(0.124395661056042, 
0.543155550956726, 0.930405616760254, 0.176615670323372, 0.122252210974693, 
1.86410081386566, 0.201039269566536, 0.00215102708898485, 0.00524011626839638, 
0.0221506990492344, 1.75632297992706, 0.954743504524231, 0.373224049806595, 
0.0127956680953503, 0.0007417316082865, 0.0123716788366437, 0.279229581356049, 
2.30779552459717, 2.58910322189331, 1.23243260383606, 0.819948613643646, 
1.74025285243988, 4.03071403503418, 2.78268098831177, 2.00978517532349, 
0.700970351696014, 0.196071043610573, 2.19463133811951, 4.83159875869751, 
2.20620393753052, 0.321354597806931, 0.00308413081802428, 1.737912774086, 
0.468539208173752, 0.0156131321564317, 0.00116395147051662, 0.0145542966201901, 
0.000892410753294826, 0.0419198162853718, 2.84171080589294, 3.22121715545654, 
2.73401832580566, 2.47091150283813, 2.10038590431213, 1.15651941299438, 
0.490403175354004, 0.0419915802776814, 0.101970501244068, 0.00181114906445146, 
0.0132269319146872, 0.212756171822548, 0.111757233738899, 1.2169703245163, 
0.129767879843712, 0, 0.582266986370087, 2.96843385696411, 1.16728830337524, 
0.0494964420795441, 0.0664984136819839), nir = c(0.261590600013733, 
0.250058531761169, 0.238313049077988, 0.246726274490356, 0.241509333252907, 
0.215491861104965, 0.25552836060524, 0.26755028963089, 0.283316373825073, 
0.2645283639431, 0.2347122579813, 0.250579416751862, 0.272739976644516, 
0.26601967215538, 0.260071456432343, 0.283827364444733, 0.270996034145355, 
0.229571804404259, 0.228905484080315, 0.240774929523468, 0.22843000292778, 
0.201068416237831, 0.174168020486832, 0.187955036759377, 0.235188364982605, 
0.226306527853012, 0.197943985462189, 0.192345812916756, 0.18694880604744, 
0.203041225671768, 0.24348683655262, 0.264572501182556, 0.234625786542892, 
0.252681404352188, 0.252072751522064, 0.241365790367126, 0.228045880794525, 
0.252986639738083, 0.261032313108444, 0.233464851975441, 0.235829710960388, 
0.235184907913208, 0.212146639823914, 0.204127430915833, 0.216947212815285, 
0.225598230957985, 0.231632620096207, 0.224976778030396, 0.219116434454918, 
0.255260914564133, 0.241265594959259, 0.237798929214478, 0.241482153534889, 
0.240964710712433, 0.252938002347946, 0.258243441581726, 0.211435839533806, 
0.217503502964973, 0.237074509263039, 0.237700119614601)), row.names = c(NA, 
60L), class = "data.frame")

[nirgrahamuk]

collect_predictions() is to get the predictions that were made on the resampled/cross validating training step, so indeed this data is different than simply your training data; therefore use the conventional predict() method with newdata= set to your training data, to simply score it up like you would any new dataset you would want to get predictions on.

[nikos_geo]

So when I'm trying to use the predict function

p <- predict(drought_res, drought_sf[, 1:4]) # excluding the column *geomtry*

I am getting the following error:

Error in UseMethod("predict") :
no applicable method for 'predict' applied to an object of class "c('resample_results', 'tune_results', 'tbl_df', 'tbl', 'data.frame')"

I even try something like
p <- predict(drought_res, drought_sf)

and more.. Can you help me type it correctly?

[nirgrahamuk]

hmmm. I think we need to go back a step. fit_resamples didnt give a model per say; it gave you multiple models.
Is there a particular one you want to use ? or do you need an alternative to fit_resamples that will get you exactly one model out of it ?

[nikos_geo]

I think I need the one with the highest R-squared. I say I think because when I build a model without spatial resampling (i.e., random resampling or stratified) and I create a random grid to tune the hyperparameters, at some point I select the model with the highest R-squared.

I was expecting something similar (?).