[MITgcm-support] "multiple tests" for testreport
Alistair Adcroft
adcroft at MIT.EDU
Mon Mar 15 09:40:34 EST 2004
I agree that we should allow code variants but I think we should limit the
code variants to not changing the "total" SIZE of a configuration. Variants
should allow for alternate schemes, forcing or tiling but if we allow the
total number of grid-points to change then the meaningful association of
files under an experiment directory becomes tenuous, even if labelled
appropriately with sub-directory names.
Why bother keeping input/ and input_alt/ and not just replace input/ with
have input/base/?
A.
--
Dr Alistair Adcroft http://www.mit.edu/~adcroft
MIT Climate Modeling Initiative tel: (617) 253-5938
EAPS 54-1523, 77 Massachusetts Ave, Cambridge, MA, USA
-----Original Message-----
From: mitgcm-support-bounces at mitgcm.org
[mailto:mitgcm-support-bounces at mitgcm.org] On Behalf Of Chris Hill
Sent: Sunday, March 14, 2004 11:54 AM
To: mitgcm-support at mitgcm.org
Subject: RE: [MITgcm-support] "multiple tests" for testreport
All,
Sounds good - but we should be a little more structured:
We should use a tree structure for this otherwise it will get illegible and
like Dimitris I want code variants too. The final structure should be
something like.
input/
input_alt/
input_alt/sea_mount_bathy
input_alt/shelf_bathy
input_alt/complex_forcing
code/
code_alt/
code_alt/s192t_8x4
code_alt/s176t_8x4
code_alt/sea_mount_bathy
results/
results_alt/sea_mount_bathy
results_alt/s192t_8x4/
If we don't do this at the top of an experiment level we will end up with
too many directories at the top and it will look messy.
**NOTE** if we allow code_alt and input_alt arbitrarily together we will
have a combinatorial explostion! The combinations exercised would therefore
be limited to {"matching_NAME_pair:matching_NAME_pair", "nomatch_NAME:base",
"base:nomatch_NAME"}, where NAME is a subdirectory entry from
[input_alt|results_alt] and "a:b" are selections from input_alt:code_alt
respectively. This means that in results_alt/ we only need one name.
Chris
> -----Original Message-----
> From: mitgcm-support-bounces at mitgcm.org
> [mailto:mitgcm-support-bounces at mitgcm.org] On Behalf Of Ed Hill
> Sent: Thursday, March 11, 2004 10:14 PM
> To: MITgcm-support
> Subject: [MITgcm-support] "multiple tests" for testreport
>
>
> Hi folks,
>
> Based on discussions with JMC and AJA, I've modified testreport so
> that one can create multiple (similar) tests for each directory within
> verification. The idea is that input files can be shared between
> similar tests.
>
> The way it works is that directories within verification will have
> multiple tests run if they meet the following criteria:
>
> 1) the test dir must contain a "build" directory
>
> 2) the test dir must contain one or more extra "input"
> directories each with names of the format "input.*"
> (eg. input.alt_bathy, input.alt_forcing, etc.)
>
> 3) the test dir must contain one or more extra "output"
> files in the format "results/output.txt.*" where the
> extension on each file matches each of the extensions
> on the input.* dir names
>
> So an example would be:
>
> exp0/input <== standard/shared input files
> exp0/input.ex <== extra and/or modif inputs
> exp0/results/output.txt.ex <== results for the ".ex" case
>
> Please let me know if you have any ideas for improvements. JMC said he
> would soon check in some examples to demonstrate it. And attached is
> an example of the output it generates.
>
> Ed
>
>
> --
> Edward H. Hill III, PhD
> office: MIT Dept. of EAPS; Room 54-1424; 77 Massachusetts Ave.
> Cambridge, MA 02139-4307
> email: eh3 at mit.edu, ed at eh3.com
> URL: http://web.mit.edu/eh3/
> phone: 617-253-0098
> fax: 617-253-4464
>
_______________________________________________
MITgcm-support mailing list
MITgcm-support at mitgcm.org
http://dev.mitgcm.org/mailman/listinfo/mitgcm-support
More information about the MITgcm-support
mailing list