Re: Siena Computing School, Tentative Program

[Favre-Nicolin Vincent <vincefn@users.sourceforge.net>]

	Hi all,

   The proposed programme looks great to me. However I'd like to suggest two 
modifications:

1) I feel that a bit too much time is spent on general programming issues, 
almost 3 days out of 5. I think it would be better to add at least half a day 
to spend on general talk about algorithms.
   To gain time:
	- On day 1, the 'procedural programming' could be merged with the 
'Object-oriented programming' (OOP). IMHO presenting procedural programming 
should serve as an introduction to OOP, but not much more (yes, I have my 
prejudices as well). (note: along with the introduction to OOP, the use of 
UML (Unified Modeling Language) could be introduced as well, although I have 
not personnaly used it)
        - also, the use of scripting languages could be reduced to python. 
csh, bash, perl (and awk) are extremely useful but are not specific to 
scientific applications, so there would be little that can be said that could 
not be found in any textbook or google. Spending more time on python which is 
becoming an outstanding scientific programming language could be better

   1/2 day (on day 3 after compression of days 123 timetable) may be spent on 
algorithms, for a general audience, from least Squares to MaxLikelihood (see 
(2) below), and also to other general algorithms (Monte-Carlo, genetic 
algorithms).

(2) LeastSquares (LS) vs MaximumLikelihood (ML) is already on day 4, but I am 
afraid that so late in the timetable it will be with a split audience and 
mostly for macromolecular crystallographers. ML is well established for 
protein crystallography, but it is _very_ slow to be used for small/molecule 
& inorganic materials.
   I strongly believe that it would be a significant achievement if _all_ 
attendants would be aware after the school that there is something beyond LS 
refinement, e.g. that (i) statistical errors are not only present in the 
experimental data, but also in the model (missing atoms, out to handle phase 
errors,...), (ii) that non-gaussian distributions should also be considered, 
etc...
   The above may seem trivial to experienced programmers who have a thorough 
knowledge of literature, but for young crystallographers, there is generally 
little said of ML and Bayesian approaches, at least outside the protein 
crystallography world. It need not be very abstract, the advantages of ML can 
be demonstrated with a linear fit of scattered point with a few outliers.

   This could be put in the half-day on algorithms, or maybe better there 
could be an entire half-day devoted to LS vs ML and statistical analysis 
(could be grouped with direct methods ?)

	Best regards,
-- 
Vincent Favre-Nicolin
Université Joseph Fourier
http://v.favrenicolin.free.fr
ObjCryst & Fox : http://objcryst.sourceforge.net

_______________________________________________
compcomm mailing list
compcomm@iucr.org
http://scripts.iucr.org/mailman/listinfo/compcomm