CCL: COMFA q² SUMMARY
- From: Antoine FORTUNE
<Antoine.Fortune]^[ujf-grenoble.fr>
- Subject: CCL: COMFA q² SUMMARY
- Date: Wed, 02 Nov 2005 15:04:52 +0100
Many thank's to all who help me and the CCL. Here is a summery of the
answers I got.
"q-sq" is the same as "crossvalidated r-sq". When you do a
comfa run
with crossvalidation (choose the leave-one-out radio button in the setup
dialog box), the r-sq reported in the summary output is the
crossvalidated r-sq. You have to do a second run with the proper number
of components and without crossvalidation to get the conventional r-sq.
Steve Bowlus
Seascape Learning, LLC (Tripos distributor)
Sent to CCL by: "Barbault Florent"
[florent.barbault(a)paris7.jussieu.fr]
When you do a cross-validation the r² in SYBYL represent the
q². So
you have to do a cross-validation analysis first, and if it is correct
a non-cross-validation analysis.
Best regards
Sent to CCL by: Prashanth Athri [athriprashanth!=!yahoo.com]
Hi,
On the Partial Least Squares Analysis dialog box:
1. Click on the SAMPLS button and set the number of
components to a value higher than what you would
expect the dimensionality of your data to be. Then
click on "Run PLS".
2. A dialog informing you about SAMPLS pops up- click
OK.
3. On the terminal window, you will eventually see
crossvalidated r-squared [alias q-squared]. The number
of components you will choose for the final run will
be the one for which the q-squared is the highest.
Hope this helps.
-Prashanth
--- "Antoine FORTUNE
Antoine.Fortune^_^ujf-grenoble.fr"
<owner-chemistry!^!ccl.net> wrote:
Sent to CCL by: Antoine FORTUNE [Antoine.Fortune-
-ujf-grenoble.fr]
Hi all,
I'm working with sybyl MSS to do comfa study. As I
run a PLS, sybyl
gives r² value in the report. Where / how can i get
the q² value ? It's
not in the report log.
Thank's
Antoine
begin:vcard
fn;quoted-printable:Antoine Fortun=C3=A9
n;quoted-printable:Fortun=C3=A9;Antoine
org:UFR Pharmacie - Universite Grenoble 1;Departement de Pharmacochimie
Moleculaire
adr:38240;;5 avenue de Verdun ;MEYLAN;;;FRANCE
email;internet:Antoine.Fortune(_)ujf-grenoble.fr
title:Ingenieur Modelisation Moleculaire
tel;work:+33 4 76 04 10 38
x-mozilla-html:TRUE
url:http://dpm.ujf-grenoble.Fr
version:2.1
end:vcard