Multi-unit Van Rossum spike train metric. This is a kernel-based implementation with markage vector and precomputed exponential factor, as described in Houghton and Kreuz, 2012, 'On the efficient calculation of Van Rossum distances.'. This package started out as a Python wrapping of the original C++ implementation given by the authors of the paper, and evolved from there with bugfixes and improvements.
Full documentation is hosted at http://pymuvr.readthedocs.org/
- Python 2.7 or 3.x.
- NumPy>=1.7.
- C++ development tools and Standard Library (package build-essential on Debian).
- Python development tools (package python-dev on Debian).
To install the latest release, run:
pip install pymuvr
If you prefer installing from git, use:
git clone https://github.com/epiasini/pymuvr.git cd pymuvr python setup.py install
Note that you'll get testing and benchbark scripts only if you install manually (i.e. not via pip).
The module exposes two functions:
pymuvr.distance_matrix(parallel_trains_1, parallel_trains_2, cos, tau)
pymuvr.square_distance_matrix(parallel_trains, cos, tau)
distance_matrix calculates the 'bipartite' (rectangular) dissimilarity matrix between the trains in parallel_trains_1 and those in parallel_trains_2.
square_distance_matrix calculates the 'all-to-all' dissimilarity matrix between each pair of trains in parallel_trains. It's an optimised form of distance_matrix(parallel_trains, parallel_trains, cos, tau).
They both return their results as a 2D numpy array.
The parallel_trains arguments must be thrice-nested lists of spiketimes, in such a way that parallel_trains[i][j][k] represents the time of the kth spike of the jth cell of the ith train. cos and tau are the usual parameters for the multiunit Van Rossum metric.
See examples/benchmark_versus_spykeutils.py for an example of usage comparing the performance of pymuvr with the pure Python implementation of the multiunit Van Rossum distance in spykeutils.
This package is licensed under version 3 of the GPL or any later version. See COPYING for details.