Discourage using bbs(, constraint "!= "none")

(solves #36)

R/bl.R

@@ -503,6 +503,12 @@ bbs <- function(..., by = NULL, index = NULL, knots = 20, boundary.knots = NULL,
 
     cll <- match.call()
     cll[[1]] <- as.name("bbs")
+
+    constraint <- match.arg(constraint)
+    if (constraint != "none")
+        warning("Using ", sQuote('bbs()'), ' with constraint != "none" is discouraged. Preferably use ', 
+                sQuote('bmono()'), " instead.\n",
+                "See section ", sQuote("Details"), " of ?bbs for more information.")
 
     mf <- list(...)
     if (length(mf) == 1 && ((is.matrix(mf[[1]]) || is.data.frame(mf[[1]])) &&

man/baselearners.Rd

@@ -170,11 +170,14 @@ bl1 \%O\% bl2
                   obtained using \code{\link[party]{ctree_control}}.
                   Defines hyper-parameters for the trees which are used as base-learners,
                   stumps are fitted by default.}
-  \item{constraint}{ type of constraint to be used. The constraint can
-    be either monotonic "increasing" (default), "decreasing" or "convex"
-    or "concave". Additionally, "none" can be used to specify
-    unconstrained P-splines. This is especially of interest in
-    conjunction with \code{boundary.constraints = TRUE}.}
+  \item{constraint}{type of constraint to be used. For \code{bmono}, 
+    the constraint can be either monotonic \code{"increasing"} (default), 
+    \code{"decreasing"}, or \code{"convex"} or \code{"concave"}. 
+    Additionally, \code{"none"} can  be used to specify unconstrained P-splines. 
+    This is especially of interest in conjunction with \code{boundary.constraints = TRUE}.
+    For \code{bbs}, the constraint can be \code{"none"}, monotonic \code{"increasing"}, or
+    \code{"decreasing"}. In general it is advisable to use \code{bmono} to fit 
+    monotonic splines.}
   \item{type}{
     determines how the constrained least squares problem should be
     solved. If \code{type = "quad.prog"}, a numeric quadratic
@@ -263,7 +266,7 @@ bl1 \%O\% bl2
   \code{boundary.knots = c(0, 2 * pi)} for angles as in a sine function
   or \code{boundary.knots = c(0, 24)} for hours during the day). For
   details on cylcic splines in the context of boosting see Hofner et
-  al. (2014).
+  al. (2014). 
 
   \code{bspatial} implements bivariate tensor product P-splines for the
   estimation of either spatial effects or interaction surfaces. Note
@@ -393,7 +396,9 @@ bl1 \%O\% bl2
   Mueller and Hothorn (2011b). The quadratic-programming based algorithm
   is described in Hofner et al. (2014). Alternative monotonicity
   constraints are implemented via T-splines in \code{bbs()} (Beliakov,
-  2000).
+  2000). In general it is advisable to use \code{bmono} to fit monotonic splines 
+  as T-splines show undesirable behaviour if the observed data deviates 
+  from monotonicty.
 
   Two or more linear base-learners can be joined using \code{\%+\%}. A
   tensor product of two or more linear base-learners is returned by