CODEWITHSUNDEEP
rtidyrgam
Samuel Chan
Samuel Chan

Reputation: 3

How do you do predict.gam on many models nested within tibbles?

I've been following https://r4ds.had.co.nz/many-models.html to create a tibble with individual GAMs for each row. The data column lists the data that was used to generate the GAM. Now I've having trouble trying to use the predict.gam function to generate a new column with the predicted values.

# A tibble: 2,157 x 3
# Groups:   Site [2,157]
   Site                     data             model 
   <fct>                    <list>           <list>
 1 Abana Rock North 1       <tibble [7 x 6]> <gam> 
 2 Abana Rock North 2       <tibble [7 x 6]> <gam> 
 3 Abana Rock South 1       <tibble [7 x 6]> <gam> 
 4 Abana Rock South 2       <tibble [7 x 6]> <gam> 
 5 Ampa Marker East         <tibble [7 x 6]> <gam> 
 6 Ampa Marker West         <tibble [7 x 6]> <gam> 
 7 Ampa Patches Southwest 1 <tibble [7 x 6]> <gam> 
 8 Ampa Patches Southwest 2 <tibble [7 x 6]> <gam> 
 9 Brunei Patches 1         <tibble [7 x 6]> <gam> 
10 Brunei Patches 2         <tibble [7 x 6]> <gam> 
# ... with 2,147 more rows

# A tibble: 7 x 6
  Country Location       Year Population100km
  <fct>   <fct>          <dbl>           <int>
1 Brunei  Inshore Brunei  1990          431102
2 Brunei  Inshore Brunei  1995          492958
3 Brunei  Inshore Brunei  2000          545008
4 Brunei  Inshore Brunei  2005          602691
5 Brunei  Inshore Brunei  2010          660197
6 Brunei  Inshore Brunei  2015          715266
7 Brunei  Inshore Brunei  2020          766133

The code thus far are as follows:

data <- rawdata %>% 
  group_by(Site) %>% 
  nest()

model_function <- function(df) {
  gam(Population100km ~ Year, data = df)
}

models <- data %>% 
  mutate(model = map(data, mode_function))

years <- data.frame(Year=1990:2020)

Now I'm basically trying to run the following for each model and save it as another column using mutate.

predict.gam(models$model, predict.years)

Any help would be appreciated. Thank you!

Upvotes: 0

Views: 511

Answers (1)

Athanasia Mowinckel
Athanasia Mowinckel

Reputation: 462

You should be able to use similar syntax for the predictions. I have here used gapminder to make it reproducible.

library(tidyverse)
library(mgcv)
#> Loading required package: nlme
#> 
#> Attaching package: 'nlme'
#> The following object is masked from 'package:dplyr':
#> 
#>     collapse
#> This is mgcv 1.8-31. For overview type 'help("mgcv-package")'.

rawdata <- gapminder::gapminder

data <- rawdata %>% 
  group_by(country) %>% 
  nest()

years <- data.frame(year=1990:2020)

models <- data %>% 
  mutate(
    model = map(data, ~ gam(lifeExp ~ year, data = .x)),
    predicted = map(model, ~ predict(.x, newdata = years))
    )

unnest(models, predicted)
#> # A tibble: 4,402 x 4
#> # Groups:   country [142]
#>    country     data              model  predicted
#>    <fct>       <list>            <list>     <dbl>
#>  1 Afghanistan <tibble [12 × 5]> <gam>       40.4
#>  2 Afghanistan <tibble [12 × 5]> <gam>       40.6
#>  3 Afghanistan <tibble [12 × 5]> <gam>       40.9
#>  4 Afghanistan <tibble [12 × 5]> <gam>       41.2
#>  5 Afghanistan <tibble [12 × 5]> <gam>       41.5
#>  6 Afghanistan <tibble [12 × 5]> <gam>       41.7
#>  7 Afghanistan <tibble [12 × 5]> <gam>       42.0
#>  8 Afghanistan <tibble [12 × 5]> <gam>       42.3
#>  9 Afghanistan <tibble [12 × 5]> <gam>       42.6
#> 10 Afghanistan <tibble [12 × 5]> <gam>       42.8
#> # … with 4,392 more rows

Created on 2020-04-10 by the reprex package (v0.3.0)

Upvotes: 1

Related Questions