The class "psi_func"
is used to store \(\psi \ (psi)\)
functions for M-estimation. In particular, an object of the class
contains \(\rho(x) \ (\code{rho})\), its derivative
\(\psi(x) \ (psi)\), the weight function \(\psi(x)/x\), and
first derivative of \(\psi\), Dpsi =
\(\psi'(x)\).
Objects can be created by calls of the form new("psi_func", ...)
,
but preferably by psiFunc(...)
.
rho
:the \(\rho()\) function, an object of
class "functionX"
. This is used to formulate the
objective function; \(\rho()\) can be regarded as generalized
negative log-likelihood.
psi
:\(\psi()\) is the derivative of \(\rho\),
\(\psi(x) = \frac{d}{dx} \rho(x)\);
also of class "functionX"
.
wgt
:The weight function \(\psi(x)/x\),
of class "functionX"
.
Dpsi
:the derivative of \(\psi\),
\(Dpsi(x) = psi'(x)\); of class "functionX"
.
Dwgt
:the derivative of the weight function,
of class "functionX"
, is generated automatically if
psiFunc
constructor is used.
tDefs
:named numeric vector of tuning parameter Default values.
%%% FIXME : Replace these by *methods*
Erho
:A function of class "functionXal"
for
computing \(E[\rho(X)]\) when \(X\) is standard normal
\(\mathcal{N}(0,1)\).
Epsi2
:A function of class "functionXal"
for
computing \(E[\psi^2(X)]\) when \(X\) is standard normal.
EDpsi
:A function of class "functionXal"
for
computing \(E[\psi'(X)]\) when \(X\) is standard normal.
name
:Name of \(\psi\)-function used for printing.
xtras
:Potentially further information.
Currently, only chgDefaults()
, plot()
and show()
.
Martin Maechler
psiFunc
.
str(huberPsi, give.attr = FALSE)
plot(hampelPsi)# calling the plot method (nicely showing "all" !)
Run the code above in your browser using DataLab