Skip to contents

Fit a MaGP model with a two-dimensional sequence map

Source: R/engine-rcpp.R
magp2d_fit.Rd

Fits an additive Gaussian process for data that combine component amounts with a component order. Sequence positions are represented by points in a compact two-dimensional latent map.

Usage

magp2d_fit(
  X,
  y = NULL,
  q = NULL,
  tau = 0.001,
  maxeval = 500,
  xtol_rel = 1e-05,
  lb_sigma = 10,
  ub_sigma = 1000,
  lb_theta = 0.5,
  ub_theta = 1000,
  lb_delta = -1,
  ub_delta = 1,
  seed = NULL,
  n_starts = 1,
  workers = 1
)

Arguments

X

A numeric matrix or data frame. The first q input columns contain quantitative levels and the next q columns contain sequence positions. Each sequence row must be a permutation of 1:q. A column named y may be included as the response.

y

An optional numeric response vector. It may be omitted when X contains a response column named y.

q

The number of components. It is inferred from the number of input columns when omitted. The two-dimensional model requires at least three components; the full model requires at least two.

tau

A fixed nonnegative nugget variance added to the covariance diagonal.

maxeval

Maximum number of objective evaluations used by nloptr.

xtol_rel

Relative parameter tolerance used by nloptr.

lb_sigma, ub_sigma

Lower and upper bounds for the additive variance parameters.

lb_theta, ub_theta

Lower and upper bounds for the quantitative correlation parameters.

lb_delta, ub_delta

Lower and upper bounds for the mapping parameters.

seed

An optional nonnegative integer used to generate the initial parameter vectors. The first start retains the result produced by this seed when n_starts = 1.

n_starts

Number of independent parameter starts. The fitted object contains the result with the lowest objective among the converged starts.

workers

Number of local worker processes. Values greater than one use a socket cluster and are capped at two or n_starts, whichever is lower.

Value

An object of class magp2d.

Details

The covariance is a sum of component-specific terms. Each term combines the distance between two quantitative levels with the distance between their mapped sequence positions. The variance, correlation, and mapping parameters are estimated together with bounded optimization.

Quantitative columns outside [0, 1] are transformed with column-wise min-max scaling. Their training ranges are stored in the fitted object and reused for prediction. Columns already in [0, 1] are left unchanged.

When several starts are requested, each start receives a separate seed. Supplying seed makes the starts and the selected fit reproducible for both sequential and parallel execution. Start-level objective values, convergence codes, warnings, and errors are stored in fit$multistart$starts.

Examples

# \donttest{
train <- read.table(
  system.file("extdata", "example_train.txt", package = "magp"),
  header = TRUE
)
fit <- magp2d_fit(train, seed = 1, n_starts = 2)
fit
#> <magp2d model>
#>   components: 4 
#>   training runs: 31 
#>   nugget: 0.001 
#>   fitted mean: 15.0858 
#>   objective: 205.1974 
#>   optimizer: converged (status 4 )
#>   starts: 2 (best 2 )
#>   execution: sequential with 1 worker(s)
# }