AW: Box-Cox transformations
- To: mathgroup at smc.vnet.net
- Subject: [mg23511] AW: [mg23479] Box-Cox transformations
- From: Wolf Hartmut <hwolf at debis.com>
- Date: Tue, 16 May 2000 02:44:57 -0400 (EDT)
- Sender: owner-wri-mathgroup at wolfram.com
my answer is below
Von: David Alan Paul [SMTP:david_alan_paul at yahoo.com]
Gesendet am: Samstag, 13. Mai 2000 04:54
An: mathgroup at smc.vnet.net
Betreff: [mg23479] Box-Cox transformations
I'm working on implementing multivariate Box-Cox transformations to
achieve approximate multivariate normality when the original data
come
from the ArcSinh-multivariatenormal distribution.
My code works fine when I restrict myself to a particular dimension
(say,
the bivariate case). I do not know how to extend my
functions/modules to
handle data of any dimension because the
FindMinimum[-LogLikelihood,{x,x0},{y,y0},...] command requires me to
specify {var,startingpoint} for every variable - and unless I assume
a
particular dimensionality, I do not know how many of these
specifications
to include.
Dear David Paul,
what you have to do is to treat your variables and starting values
as objects of their own.
_Somewhere_ you have to specify them, e.g. as
vars = {x, y, z}; (* these must be symbols without a value *)
starts = {x0, y0, z0}; (* these are the starting points,
substitute directly or give the x0,... a value respectively *)
FindMinimum[function, ##] & @@ Transpose[{vars, starts}]
FindMinimum::"fmns": "Starting value \!\(x0\) in \!\({x, x0}\) is
not a real \
number."
(* the error was provoked to see the following: *)
FindMinimum[function, {x, x0}, {y, y0}, {z, z0}]
Kind regards, Hartmut