New cyclic_basis()
for modelling seasonal effects.
Bernstein_basis(..., ui = c("increasing", "zerointegral")
came with incorrect constraint matrix.
Bernstein_basis(..., ui = "zerointegral")
came with incorrect constraint matrix.
Allow coefficients in matrix form.
Add convexity and concavity constraints to polynomials in Bernstein form.
Add toy example demonstrating constrainted
curve estimation to ?Bernstein_basis
.
More infrastructure for as.basis
to return
a Matrix
object for ordered factors.
Update citation info.
Change name of parameter in intercept\_basis
to
(_Intercept_)
; this helps to detect
intercepts originating from other calls to model.matrix
..
Make sure bounds[1]
is positive if log_first = TRUE
.
Correct behaviour outside support when log_first = TRUE
.
model.matrix
has constrasts
attribute where appropriate.
Bernstein(log()) is possible with new log_first
argument.
Fix > length 1 issues.
Constrain Bernstein to zero second derivative at the support when linear extrapolation shall be used. This avoids ugly bumps in the corresponding densities in mlt.
Replace rBind
by rbind
as requested by MM.
A paper describing version 1.0-0 of the mlt, basefun, and variables packages was accepted for publication in the Journal of Statistical Software 2018-03-05.
Documentation updates.
nparm
sometimes generated data with sanitised variable
names, we don't want this to happen.
Allow positive increasing Bernstein polynomials.
Add intercept_basis
, ie an intercept only model matrix.
model.matrix.cbind_bases
can now deal with
lists as newdata
argument (given appropriate
dimensions in dim
).
Deal with integer variables when sumconstr = TRUE
.
Remove unnecessary warning when sumconstr = TRUE
and
negative = TRUE
.
Register C routines.
Fix bug in predict
when newdata
was a data frame.
Fix bug in predict
when newdata
was a list. The
ordering of the resulting array may have been incorrect under
rare circumstances.
Approximate Bernstein polynomial f outside its support
(l, u)
by a linear function f(u + x) = f(u) + f'(u) * (u - x), now also
for deriv > 0
(maxderiv
argument in
model.matrix.Bernstein_basis
. The default (linear extrapolation
for deriv = 0
only) did not change.
Contraints in log_basis
were potentially wrong.
log_basis
has better checks for positivity of
the variable to be log-transformed.
The basefun package was first published on CRAN.