search

rstudio / bundle

Prepare objects for serialization with a consistent interface
https://rstudio.github.io/bundle/
Other
27 stars 4 forks source link

support caret #19

Closed simonpcouch closed 2 years ago

simonpcouch commented 2 years ago

Just documenting a quick look into whether we need to support caret. From my understanding, caret handles serialization by serializing objects that need it and, at predict time, temporarily unserializing them.

library(caret)
#> Loading required package: ggplot2
#> Loading required package: lattice

predictors <- mtcars[, c("cyl", "disp", "hp")]

set.seed(1)

suppressMessages(suppressWarnings(
  fit <-
    train(
      x = predictors,
      y = mtcars$mpg,
      method = "mlpKerasDecay",
      verbose = 0
    )
))

callr::r(
  function(fit, predictors) {
    library(caret)

    predict(fit, predictors)
  },
  args = list(fit = fit, predictors = predictors)
)
#>  [1]  -38.6654625  -38.6654625  -17.0129700  -92.0914993 -117.8382492
#>  [6]  -76.5140152  -84.0575333  -53.0708122  -33.9291649  -36.5351295
#> [11]  -36.5351295  -69.5225601  -69.5225601  -69.5225601 -164.4191284
#> [16] -153.0513000 -134.9093323  -14.0693846  -19.1900330  -10.4087210
#> [21]  -21.6791229 -107.0059433  -99.3736343  -78.6058884 -139.6448059
#> [26]  -14.2329397  -24.6836357   -0.3287155  -69.9819717    0.8798035
#> [31]   -8.4605131  -16.3787842

Created on 2022-07-15 by the reprex package (v2.0.1)

So, good to go. :)

simonpcouch commented 2 years ago

With a train object, x$finalModel would be the slot serialize.