Re. Performances

From: Jean Thioulouse (Jean.Thioulouse@biomserv.univ-lyon1.fr)
Date: Wed Sep 17 1997 - 19:00:47 MET DST


Federico Spinazzi <federico@SysPr03.disat.unimi.it>
>1) Why are the computation time so different (the new version of
>distance ask for more RAM also, under WinNT) ? Why are they worst ?

Sorry I have no idea about this. I have not changed the algorithm
in the PCO and I have not changed the compilation options. The figures
you give for computation times are consistent with the results I have
obtained.

>2) One can select the distance type (Euclidean, Chi2, Jaccard) in the
>option 'Distances|Canonical Distance' no more, why ?

The Chi2 distance can be computed with the "Triplet to Distance" option,
and the Jaccard distance can be computed with the "Binary Dissimilarity"
option (with coefficient number 1). You can also use the Clusters module.

>I think also that if you ask for a column distance matrix and THEN for a
>row distance matrix, from the same file and with the same file as output,
>you will always obtain a column distance matrix unless you restart the module.

Yes, this is a bug. I am going to fix it soon, but I'll have to update
all the modules (sigh !).

Jean

--
         Jean Thioulouse - Laboratoire de Bio-Pedologie
               ORSTOM - BP 1386 - Dakar - Senegal
Fax: (221) 32 16 75 - Tel: Office: (221) 32 18 46 / Home: (221) 32 36 33
           http://biomserv.univ-lyon1.fr/JTHome.html



This archive was generated by hypermail 2b30 : Sat Feb 10 2001 - 10:21:36 MET