search

boost-R / mboost

Boosting algorithms for fitting generalized linear, additive and interaction models to potentially high-dimensional data. The current relase version can be found on CRAN (http://cran.r-project.org/package=mboost).
73 stars 27 forks source link

Fix monotonic interactions with ordinal variables #73

Open hofnerb opened 7 years ago

hofnerb commented 7 years ago 
## monotonicity with interaction and factor
library("mboost")

## simulate some data
set.seed(1234)
x1 <- rnorm(1000)
x2 <- sample(1:5, 1000, replace = TRUE)
y <- -x1^3 * (6 - x2) + rnorm(1000, sd = 0.1)
x2 <- as.factor(x2)

## Works with numeric values
x2 <- as.numeric(x2)
mod <- mboost(y ~ bmono(x1, x2, df = 5, constraint = list("decreasing", "increasing")), 
              control = boost_control(trace = TRUE, mstop = 10))
plot(mod)

## Interaction currently does not work with ordinals
x2 <- as.ordered(x2)

## Does work as single effect
mod2 <- mboost(y ~ bmono(x2, constraint = "decreasing"))
plot(mod2)

## Doesn't work
mod3 <- mboost(y ~ bmono(x1, x2, df = 5, constraint = list("decreasing", "increasing")), 
               control = boost_control(trace = TRUE, mstop = 10))
##  Error in seq.default(boundary.knots[1] - degree * dx, boundary.knots[1],  : 
##  'from' cannot be NA, NaN or infinite

Of note:

x1 <- as.ordered(x1)
x2 <- as.ordered(x2)
mod <- mboost(y ~ bmono(x1, x2, constraint = list("decreasing", "increasing")))
##  Error in bl_mono(ret, Xfun = X_ols, args = args) : 
##  Bivariate monotonic effects currently not implemented for ordered factors