mailrelax version 1.3.14.

Others Months | Index by Date | Thread Index
>>   [Date Prev] [Date Next] [Thread Prev] [Thread Next]



Posted by Edward d'Auvergne on March 20, 2012 - 15:07:
This is a major bug fix release focusing on memory usage, performance,
and IO.  The parallised model-free optimisation code (for running on
clusters/grids using the multi-processor package) was causing relax to
crash on 32-bit systems due to the system running out of virtual
memory.  The new version should only require about a third of the
memory of relax 1.3.13 (see  The
parts of relax dealing with atomic or spin specific data (reading of
spin systems from PDB files, reading and writing of spin specific
data, etc.) have been rewritten for efficiency and should now be many
orders of magnitude faster for certain work loads.  There have also
been large improvements to the multi-processor framework, specifically
the creation of a public API and a reference implementation.  In
addition there are improvements for IO capture on both the
uni-processor and mpi4py-processor fabrics for relaying slave messages
back to the master, better integration with the GUI and compatibility
with the logging and teeing command line flags.  Other changes include
expansion of the relax information print out, support for numpy data
structures in save files, and the addition of the rdc.delete and
pcs.delete user functions.  Feel free to upgrade if you would like the
benefits of this new version.

The new relax versions can be downloaded from either or  If binary distributions are not yet
available for your platform and you manage to compile the binary
modules, please consider contributing these to the relax project
(described in section 3.6 of the relax manual,

The full list of changes is:

        * An expanded system information printout (from 'relax -i').
        * Improvements to the multi-processor implementation,
including creation of a public API and a reference implementation.
        * Overhaul of the multi-processor stream capture design - the
slave IO streams are now properly redirected to the master processor
and the master no longer overrides stream capture by the relax user
        * Numpy data structures are now supported in relax save files.
        * Implemented the rdc.delete and pcs.delete user functions.
        * Speed ups of a number of user functions involved in looping
over atomic or spin specific data - this can be significant if a lot
of structural operations are performed, or a large number of
structures are required.
        * Redesigned how spin parameters are handled in the specific
analyses (which includes parameter name changes).

        * The MS Windows text wrapping width has been changed from 80
to 79 - this now fits the cmd prompt.
        * The program intro print out and prompt UI help system is
using the new Status.text_width variable to wrap text.
        * The model-free constraint matrix A is now of numpy.int8 type
- this is to decrease virtual memory usage and increase the scaling
efficiency on clusters.
        * Changed all relax website links to
for consistency.
        * The script can now be imported for the epydoc API
documentation system.
        * The extern and minfx.scipy_subset relax packages are
excluded form the API documentation.
        * Added the ability to skip scripts in the package __all__
list unit test checks.
        * Added some code to detect the bit version of MS Windows in
the information print out (to better distinguish 32 vs. 64-bit
        * Added some documentation about master processors on Linux
2.6 eating 100% of one CPU core (in the mpi4py multi-processor
        * Added a reference implementation to the multi-processor
package - this is to demonstrate to users of the package how an
implementation is created via the public API.
        * Elimination of all relax dependencies from the
multi-processor package.
        * Added support for the memory size on MS Windows to the relax
info print out.
        * Updated the value.set user function unknown parameter error
message to list the known ones.
        * For Unix and GNU/Linux systems, the relax info printout now
shows the ram and swap size.
        * Expansion and improvement of the information printed out by
'relax --info'.
        * Expansion of the multi-processor API documentation.
        * Expansion of the multi-processor package documentation with
a step by step usage guide - this should increase the usability of the
package by clarifying how one should use it.
        * Expansion and improvements to the multi-processor package
and module docstrings.
        * The finish button in the new analysis wizard has been
renamed to "Start".
        * Created a special Verbosity singleton for controlling the
multi-processor package print outs.
        * Future proofed the relax codebase by replacing all ''' with
""" in the docstrings.
        * Removed the unnecessary try statement in the model-free method as exception handling is correctly
performed on the slave and master.
        * Shifted the Memo object into its own module (multi package).
        * Simplification and abstraction of the
method to shift all exception handling into the package.  Therefore
program code no longer needs to handle the multi-processor specific
        * Created a new module 'multi.misc' for holding miscellaneous
functions used throughout the multi package.
        * Created a public API for the multi-processor package,
available via multi.__init__.
        * The load_multiprocessor() function is no longer a static
method of the Processor base class.  This function loads the correct
Processor class, so doesn't need to be a method of the base class and
operates cleanly and more clearly as a stand alone function.
        * Clean up of the processor IO module (multi package).
        * Eliminated all usage of sys.__stdout__ and sys.__stderr__ in
the multi-processor package - this returns full control of IO streams
to the parent program.
        * The float arg checks now checks against all the numpy float
types (float16, float32, float64, float128).
        * Added value.write user function calls to the J(w) mapping
system test script.
        * Added some value.write user functions to the J(w) mapping
sample script.
        * Modified the J(w) mapping test data to include relaxation
values of None to trigger bug #19329 (
        * Speed up for the function.
        * More speed ups for the
Selection.__contains_mol_res_spin_containers() method.
        * Reordered the checks in
Selection.__contains_mol_res_spin_containers() - this cuts the number
of function calls down by avoiding calls if residue
or spin numbers match.
        * Simplified the function - this
is to minimise the number of isinstance() calls when dealing with the
relax mol-res-spin sequence data.
        * Updates for Python 3.0 using 2to3.
        * Removal of a number of debugging print out statements.
        * Significant speed ups of the return_spin() and
return_residue() functions.
        * Added a print out for the diffusion_tensor.init user
function to inform the user of an angle unit change - this is in
response to bug #19323 to make it clearer that a parameter conversion
has occurred.
        * Created a special specific API object called SPIN_PARAMS -
this will be used to handle all operations to do with model
parameters.  The object Param_list has methods for parameter
initialisation and handling (where all info is specified such as Grace
string, units, default value, etc) and for determining if a parameter
        * Mass conversion to the new GLOBAL_PARAMS and SPIN_PARAMS
specific API data structures.  The parameters are now all lowercase,
for example ['S2', 'te', 'Rex'] is now ['s2', 'te', 'rex'].  The
follow parameters are now converted throughout relax: 'bond_length' to
'r', 'CSA' to 'csa', 'heteronucleus' to 'heteronuc_type', 'proton' to
        * Created a new algorithm for finding the pivot of motion
between different structural models - this is available through the
structure.find_pivot user function.
        * Added the validate_models() method to the structural API -
this is used to check that the models are 100% consistent.
        * Added the centroid argument to the structure.superimpose
user function - this allows for the superimposition of structures
assuming a pivoted motion.

        * Fix for the title of the analysis selection wizard GUI element.
        * Fix for the MemoryError during the "final" run of d'Auvergne
Protocol affecting 32-bit systems (bug #19528,
        * Epydoc docstring fixes for the API documentation.
        * Bug fix for old numpy versions missing the float16 object.
        * The relax prompt mode now prints the intro message to
sys.stdout - this fixes this mode with the logging option or in the
mpi4py multiprocessor mode.
        * Fix for the storage of numpy data structures in the relax
save files - if special structures are created by the user, these were
not correctly stored.  The precision was often too low (printed at the
numpy default of a precision of 8), and if one dimension had more than
1000 elements, ellipses would be stored instead (the ellipses would
cause the save file to be unreadable by relax).
        * Created the SplitIO.flush() method so that a flush() call
now works with the -t and -l command line args - this simply calls the
flush() methods of both streams.
        * The relax data store __repr__() method now prints out all
objects within the base object.
        * Fix for a rare bug in the model-free analysis for when
relaxation data of None is encountered.
        * Fixes for the Dasha system test for the bug #19332 fix
        * Fixes for the Dasha use functions for file creation and data
extraction - this should now handle relaxation data values of None,
and also set the parameter values of deselected spins using the Dasha
output files.
        * Big bugfix for the user function - the wrong spin
ID string was being used to set the parameter value so that all spins
have the same value set for the parameter(bug #19332,
        * Bug fix for the consistency testing analysis for when
relaxation data of None is encountered.
        * The spin is now deselected if there is really less than
three data points, excluding data points of None (bug #19329,
        * Bug fix for the arg_check.is_none() function - the
RelaxNoneError was not imported!
        * Fix for the numbering of the TER record and subsequence
HETATM records in the PDB file creation.  The structure.write_pdb user
function now creates the correct TER record after the ATOM records,
and the HETATM records are renumbered if needed.
        * Bug fix for the compilation of the user manual.  The
molecule user function class used the temporary variable i, which was
causing the fetch_docstrings script to fail.

Related Messages

Powered by MHonArc, Updated Tue Mar 20 23:00:02 2012