Add Decoy Predictors to a Surreal Dataset

This function adds predictors of pure noise to a dataset made by surreal(). The hidden image then shows only in the residuals of the model with the real predictors: leave some out and it is not there yet, and let decoys in and it blurs. Finding it becomes a variable selection problem, which surreal_path() works through step by step.

Usage

surreal_decoys(data, n = 30, sd = NULL, shuffle = TRUE)

Arguments

data

A data frame from surreal(), surreal_text() or surreal_image(), with a response named y.

n

Integer. Number of decoy predictors to add. Default is 30.

sd

Numeric or NULL. Standard deviation of the decoys. If NULL (default), the average standard deviation of the predictors in data.

shuffle

Logical. If TRUE (default), the decoys are mixed in among the real predictors and every predictor is renamed X.1, X.2, ..., so that neither position nor name gives a decoy away. If FALSE, the decoys are added after the real predictors as D.1, D.2, ....

Value

The data frame with n more columns. The names of the decoys are kept in its "decoys" attribute, which is the answer key. The attribute is not written out by write.csv().

See also

surreal_path() to select the predictors one step at a time.

Examples

set.seed(114)
hidden <- surreal(r_logo_image_data)

# Mix in 30 decoys
decoyed <- surreal_decoys(hidden)
names(decoyed)
#>  [1] "y"    "X.1"  "X.2"  "X.3"  "X.4"  "X.5"  "X.6"  "X.7"  "X.8"  "X.9" 
#> [11] "X.10" "X.11" "X.12" "X.13" "X.14" "X.15" "X.16" "X.17" "X.18" "X.19"
#> [21] "X.20" "X.21" "X.22" "X.23" "X.24" "X.25" "X.26" "X.27" "X.28" "X.29"
#> [31] "X.30" "X.31" "X.32" "X.33" "X.34" "X.35"

# The answer key
attr(decoyed, "decoys")
#>  [1] "X.1"  "X.4"  "X.5"  "X.6"  "X.7"  "X.9"  "X.10" "X.11" "X.12" "X.13"
#> [11] "X.14" "X.15" "X.17" "X.18" "X.19" "X.20" "X.21" "X.22" "X.23" "X.24"
#> [21] "X.26" "X.27" "X.28" "X.29" "X.30" "X.31" "X.32" "X.33" "X.34" "X.35"

# The model with every predictor no longer shows a clean image
model <- lm(y ~ ., data = decoyed)
plot(model$fitted, model$resid, pch = 16)