SIMCA Classification Model (parsnip)

simca() defines a soft independent modeling of class analogy (SIMCA) classification model for tidymodels: a principal component model of each class, and the assignment of each observation to the closest class. Its engine "rsimca" (the default) is the robust SIMCA of rsimca(), whose class models resist outlying spectra of the training data.

Usage

simca(
  mode = "classification",
  num_comp = NULL,
  gamma = NULL,
  squared = NULL,
  engine = "rsimca"
)

# S3 method for class 'simca'
update(
  object,
  parameters = NULL,
  num_comp = NULL,
  gamma = NULL,
  squared = NULL,
  fresh = FALSE,
  ...
)

Arguments

mode: The type of model: "classification", the only one.
num_comp: The number of components of the PCA model of each class, or NULL (default) to let the engine choose them.
gamma: The weight of the orthogonal distances in the classification rule, between 0 and 1, or NULL (default) for 0.5.
squared: A logical: combine the squared scaled distances (TRUE) or the scaled distances (FALSE) in the classification rule, or NULL (default) for TRUE.
engine: The computational engine: "rsimca" (default).
object: A simca model specification.
parameters: A one-row tibble or named list of main parameters to update, such as those returned by tune::select_best().
fresh: If TRUE, the arguments replace those of object; if FALSE (default), they update them.
...: Not used.

Value

A model specification of classes simca and model_spec.

Details

simca() is a parsnip model specification: it is fitted with parsnip::fit() or in a workflow, and can be tuned and resampled like any parsnip model. It needs the parsnip package, and the "rsimca" engine is available once parsnip and specProc are both loaded.

Main arguments

num_comp is the number of components of the robust PCA of each class (ncomp of rsimca()), the same for all the classes. With NULL (the default), each class gets the number chosen by robpca() from the proportion of variance explained.
gamma is the weight of the orthogonal distances, against the score distances, in the classification rule (gamma of rsimca()). With NULL (the default), it is 0.5.
squared chooses the classification rule (squared of rsimca()): the squared scaled distances (TRUE, rule R2 of the paper) or the scaled distances (FALSE, rule R1). With NULL (the default), it is TRUE.

All three can be tuned with tune::tune(), using dials::num_comp(), simca_gamma() and simca_squared().

Engine arguments

The other arguments of rsimca() are set with parsnip::set_engine(): kmax, alpha, var_explained, prior, ndir and nsamp.

Predictions

predict() gives the classes (type = "class"), or with type = "raw" the combined distances to the classes and whether each observation is outlying for all of them (predict.specproc_rsimca()). SIMCA does not give class probabilities: when tuning, use metrics of the predicted classes, such as yardstick::metric_set(yardstick::accuracy, yardstick::kap).

The fitted rsimca() model is returned by parsnip::extract_fit_engine(). rsimca() uses random directions and subsets, so use set.seed() before fitting for reproducible results.

Author

Christian L. Goueguel

Examples

library(parsnip)
spec <- simca(num_comp = 2, gamma = 0.5) |>
  set_engine("rsimca", alpha = 0.75)
spec
#> SIMCA Model Specification (classification)
#> 
#> Main Arguments:
#>   num_comp = 2
#>   gamma = 0.5
#> 
#> Engine-Specific Arguments:
#>   alpha = 0.75
#> 
#> Computational engine: rsimca 
#> 
# iris: train on 100 flowers, predict the 50 others
set.seed(1)
train <- sample(nrow(iris), 100)
fit <- fit(spec, Species ~ ., data = iris[train, ])
table(predict(fit, iris[-train, ])$.pred_class, iris$Species[-train])
#>             
#>              setosa versicolor virginica
#>   setosa         16          0         0
#>   versicolor      0         17         0
#>   virginica       0          2        15
head(predict(fit, iris[-train, ], type = "raw"))
#> # A tibble: 6 × 4
#>   setosa versicolor virginica outlying
#>    <dbl>      <dbl>     <dbl> <lgl>   
#> 1 0.0893       7.77      16.8 FALSE   
#> 2 0.214        6.41      14.7 FALSE   
#> 3 0.0472      10.1       19.0 FALSE   
#> 4 0.0234       8.37      17.2 FALSE   
#> 5 0.275        5.68      14.0 FALSE   
#> 6 0.0947      10.7       20.0 FALSE