-------- Original Message --------
Subject: Re: more CVA and LDF
Date: Tue, 21 Oct 2008 07:44:43 -0700 (PDT)
From: annat haber <[EMAIL PROTECTED]>
To: [email protected]
References: <[EMAIL PROTECTED]>
Hi
I just started using Claude's book "Morphometrics with R" and it's
excellent. I wish it came out a year ago - would have saved me a lot of
time and pain. It breaks down every procedure into the smallest
functions, so it provides most of the basic codes you'd need but also
leaves you enough room to practice writing your own in the way of
combining them and incorporating data manipulation.
I use lda for LDF/CVA, and it seems to work fine for any number of
groups. the predict function gives you the posterior probabilities and
allows you to test group membership for un-assigned specimens using the
argument "newdata". So it's
L <- lda(X, gr)
pred<-predict(L)
where X is a specimen-by-variable data matrix and gr is a vector
specifying the grouping for each specimen in the same order that they
appear in the matrix (so you don't need them to be sorted by groups in
the data matrix and it can be any number of groups). The argument
"method" allows you to specify which method is used. This will find
either the dicriminant function (for two groups) or the canonical
variates (for more than two groups).
I don't know of another function in R for CVA/LDF - Fabio, which other
function did you mean? If you meant the function cancor, that's for
canonical correlations not canonical variates.
Cheers,
Annat
~~~~~~~~~
Annat Haber, PhD candidate
Committee on Evolutionary Biology
University of Chicago
Culver Hall 402
1025 E. 57th St.
Chicago IL 60637
Office: Hinds 289
c: 773 576 4205
http://home.uchicago.edu/~annat/
On Tue, Oct 21, 2008 at 7:08 AM, morphmet
<[EMAIL PROTECTED]
<mailto:[EMAIL PROTECTED]>> wrote:
-------- Original Message --------
Subject: Re: more CVA and LDF
Date: Fri, 17 Oct 2008 18:15:15 -0700 (PDT)
From: Paul Sanfilippo <[EMAIL PROTECTED] <mailto:[EMAIL PROTECTED]>>
To: [email protected] <mailto:[email protected]>
References: <[EMAIL PROTECTED]
<mailto:[EMAIL PROTECTED]>>
I'm also keen to hear any discussion about the differences between
these
two methods and also how to perform particularly a CVA in R. I've used
the 'lda' function with cross-validation on a dataset which I think is
giving me the information I want, but how do you perform a CVA, Fabio
(or anyone else)?
On a side note to R (which I've been using a bit lately and am
finding I
quite like, as steep as the learning curve is), has anyone read a book
called 'Morphometrics with R' by Julien Claude? I'm mainly
interested in
learning how to use R to perform various statistical techniques
once the
data has been analysed with the various other GMM software, so I don't
know how applicable this book is for me.
Thanks,
Paul Sanfilippo
Uni Melbourne
Australia
On Fri, Oct 17, 2008 at 8:41 PM, morphmet
<[EMAIL PROTECTED]
<mailto:[EMAIL PROTECTED]>
<mailto:[EMAIL PROTECTED]
<mailto:[EMAIL PROTECTED]>>> wrote:
-------- Original Message --------
Subject: more CVA and LDF
Date: Fri, 17 Oct 2008 01:59:31 -0700 (PDT)
From: Fabio de Andrade Machado <[EMAIL PROTECTED]
<mailto:[EMAIL PROTECTED]>
<mailto:[EMAIL PROTECTED] <mailto:[EMAIL PROTECTED]>>>
To: [email protected]
<mailto:[email protected]>
<mailto:[email protected] <mailto:[email protected]>>
Hi all,
Sorry to bring this up once more, but I don´t think any response on
that have been posted (recently) on the list.
What are the key differences of Canonical Variate Analysis and
Linear
Discriminant Functions?
On Pete E. Lestrel´s "Morphometrics for the Life Science" it
reads "A
discriminant function is a linear equation, one per individual
specimen, derived from the set of original variables x1, x2, x3,
....,
xn, each of which is multiplied by a 'weighting' coefficient,
a1,a2,a3,...,an (...). A set of these discriminant functions
(...) is
then computed for each individual case. Each discriminant function
score is orthogonal with respect to all others". (p. 159-160)
About Canonical Variate Analysis he says that it is design to
calculate "[1](...) the correlation between two derived
variables and
[2] a set of canonical variates or canonical functions (as suns of
weighted variables)(...)" (p.161).
About both he says "Although displayed results are indistinguishable
in many respects, the purpose are different" (p.160), being the CVA
basically an graphical aid for LDF.
Can anyone send some references on the specifics of each method?
They
seem distinct somehow. On R there are two different functions, being
the one associated with CVA basically for graphical purposes
(which is
confusing, as the one can display graphically the results of
discriminant functions and they seem basically the same).
cheers,
--
Fabio de Andrade Machado
Laboratorio de Herpetologia/Morfometria
Museu de Zoologia da USP
Av. Nazaré, 481, Ipiranga
São Paulo, SP, 04263-000
Brazil
+55 11 61658120
+55 11 82631029
--
Replies will be sent to the list.
For more information visit http://www.morphometrics.org
<http://www.morphometrics.org/>
--
Replies will be sent to the list.
For more information visit http://www.morphometrics.org
<http://www.morphometrics.org/>
--
Annat Haber, Ph.D. candidate
Committee on Evolutionary Biology
University of Chicago
1025 E. 57th St. Culver Hall 402
Chicago IL 60637
[EMAIL PROTECTED] <mailto:[EMAIL PROTECTED]>
773 576 4205
--
Replies will be sent to the list.
For more information visit http://www.morphometrics.org
