Skip to contents

Fit a self-organizing map to a DEA problem

Source: R/compute-som.R
compute_som.Rd

Trains a hexagonal self-organizing map (SOM) on the standardised input/output data and records, for every node, the mean DEA efficiency of the DMUs mapped to it. The result feeds plot_io_som (and other SOM views), so the map is trained once and reused.

Usage

compute_som(
  x,
  rts = c("crs", "vrs"),
  xdim = 8,
  ydim = 8,
  rlen = 1000,
  seed = NULL,
  ...
)

Arguments

x

A dea_data object, or a data frame coerced by as_dea_data.

rts

Returns to scale, passed to compute_efficiency: "crs" or "vrs".

xdim, ydim

Grid dimensions (positive integers; default 8 each).

rlen

Number of training iterations (default 1000).

seed

Optional single number; if supplied, SOM training is made reproducible without permanently changing the session's random state.

...

Further arguments passed to som.

Value

An object of class dea_som: a list with the fitted som (a kohonen object), the DMU labels, per-DMU efficiency, per-node mean node_efficiency (NA for empty nodes), and the grid dimensions.

See also

plot_io_som, som

Examples

df <- data.frame(
  dmu  = paste0("D", 1:12),
  i_x1 = runif(12, 2, 9), i_x2 = runif(12, 1, 5),
  o_y1 = runif(12, 3, 9), o_y2 = runif(12, 1, 4)
)
som <- compute_som(df, xdim = 4, ydim = 4, seed = 1)
som
#> <dea_som>
#>   grid           : 4 x 4 hexagonal (16 nodes)
#>   DMUs           : 12
#>   RTS            : CRS
#>   occupied nodes : 10