 |
|
Parplex - the parallel neural network simulator |
|
This is the preliminary documentation for parplex. Currently only a
module and class reference are available.
|
Status: |
alpha
|
|
Version: |
0.3
|
|
Latest build: |
15th August 2007
|
|
Requirements: |
Python 2.3 or later. Patience.
|
|
Download: |
Windows
installer for Python 2.5 (You may need this
DLL).
Source
code (Run python setup.py install to compile
and install).
|
|
Known Bugs: |
Parallel support is currently moribund
|
The class reference is here.
Parplex changelog
Wednesday, August 15, 2007
Major performance enhancement
The new verion has a new numerical method. It is essentially based on well known alogirthm of Micheal Hines for solving branched geomteries. It includes an adaptive step size technique adapted from the thalamocortical neuron software of Karoly Antal.
The new method is orders of magnitude faster than the existing parplex methods for compartmental models. It also _seems_ to be faster than NEURON, but this may simple reflect shortcomings in the NEURON code that I am using.
Currently, there is only a C API and to make use of the method it is necassary to subclass specific Hines integrable Dynamics classes. These are available for voltage and calcium gated channels, internal calcium dynamics and linear synapses. Documentation is (err, hmm) still coming, but is a working example based on the dentate gyrus model of Santhakumar et al, J Neurophys, 2005 93:437-53, here.
Posted 12:59 PM
by evan
Friday, November 19, 2004
Minor change
The default strength for the enq method of the Dynamics class has been changed from 0 to 1.
Posted 3:47 PM
by evan
Friday, November 12, 2004
Major feature: Writing your own C language Dynamics subclasses
The major feature of the 12th November 2004 release is that user written C language Dynamics subclasses now have their own namespaces and can be loaded as seperate modules. This means that:
- new C language Dynamics names will not clash with the C language classes that I have provided (how many names are their for a Na channel?).
- C language modules do not need to be linked into the main module and can be located anywhere in the search path. This means that user C language modules can be located with other project specific files and many versions can easily coexist.
Currently coding the classes is still clumsy, but this will be improved shortly. Also there is no documementation but look at dyn_*.c in the source distribution. When your module is coded run the script, provided in the parplex installation directory
python mkdll.py srcfile [srcfile ...]
and this will build your module in the current directory.
Note: this has not been extensively test and, in particular on windows, various compiler and python combinations will present "challenges".
Contact me for help.
Posted 11:00 AM
by evan
Bugfix: C language Dynamics class not routing some methods
The class defininition in dynamics.c did export current() and derivs() across the API (ie you couldn't call them from within Python). In addition, acceptor(), deleter() and cleanup() are exported, although I can't see a reason to call these from within Python.
Posted 9:57 AM
by evan
Thursday, November 11, 2004
Bonus for Parplex chief scientist
Evan Thomas, the architect and chief engineer of parplex, was awarded a bonus for his outstanding work on parplex. Evan enjoyed his take away latte, by himself, while watching the rain.
Posted 2:36 PM
by evan
