PySpike

https://travis-ci.org/mariomulansky/PySpike.svg?branch=master

PySpike is a Python library for the numerical analysis of spike train similarity. Its core functionality is the implementation of the bivariate ISI and SPIKE distance [1] [2]. Additionally, it provides functions to compute multivariate SPIKE and ISI distances, as well as averaging and general spike train processing. All computation intensive parts are implemented in C via cython to reach a competitive performance (factor 100-200 over plain Python).

PySpike provides the same fundamental functionality as the SPIKY framework for Matlab, which additionally contains spike-train generators, more spike train distance measures and many visualization routines.

All source codes are published under the BSD_License.

Requirements and Installation

PySpike is available at Python Package Index and this is the easiest way to obtain the PySpike package. If you have pip installed, just run

sudo pip install pyspike

to install pyspike. PySpike requires numpy as minimal requirement, as well as a C compiler to generate the binaries.

Install from Github sources

You can also obtain the latest PySpike developer version from the github repository. For that, make sure you have the following Python libraries installed:

numpy
cython
matplotlib (for the examples)
nosetests (for running the tests)

In particular, make sure that cython is configured properly and able to locate a C compiler, otherwise PySpike will use the much slower Python implementations.

To install PySpike, simply download the source, e.g. from Github, and run the setup.py script:

git clone https://github.com/mariomulansky/PySpike.git
cd PySpike
python setup.py build_ext --inplace

Then you can run the tests using the nosetests test framework:

nosetests

Finally, you should make PySpike's installation folder known to Python to be able to import pyspike in your own projects. Therefore, add your /path/to/PySpike to the $PYTHONPATH environment variable.

Spike trains

In PySpike, spike trains are represented by one-dimensional numpy arrays containing the sequence of spike times as double values. The following code creates such a spike train with some arbitrary spike times:

import numpy as np

spike_train = np.array([0.1, 0.3, 0.45, 0.6, 0.9])

Loading from text files

Typically, spike train data is loaded into PySpike from data files. The most straight-forward data files are text files where each line represents one spike train given as an sequence of spike times. An exemplary file with several spike trains is PySpike_testdata.txt. To quickly obtain spike trains from such files, PySpike provides the function load_spike_trains_from_txt.

import numpy as np
import pyspike as spk

spike_trains = spk.load_spike_trains_from_txt("SPIKY_testdata.txt",
                                              time_interval=(0, 4000))

This function expects the name of the data file as first parameter. Additionally, the time interval of the spike train measurement can be provided as a pair of start- and end-time values. If the time interval is provided (time_interval is not None), auxiliary spikes at the start- and end-time of the interval are added to the spike trains. Furthermore, the spike trains are sorted via np.sort (disable this feature by providing is_sorted=True as a parameter to the load function). As result, load_spike_trains_from_txt returns a list of arrays containing the spike trains in the text file.

If you load spike trains yourself, i.e. from data files with different structure, you can use the helper function add_auxiliary_spikes to add the auxiliary spikes at the beginning and end of the observation interval. Both the ISI and the SPIKE distance computation require the presence of auxiliary spikes, so make sure you have those in your spike trains:

spike_train = spk.add_auxiliary_spikes(spike_train, (T_start, T_end))
# if you provide only a single value, it is interpreted as T_end, while T_start=0
spike_train = spk.add_auxiliary_spikes(spike_train, T_end)

Computing bivariate distances profiles

Important note:

Spike trains are expected to be sorted! For performance reasons, the PySpike distance functions do not check if the spike trains provided are indeed sorted. Make sure that all your spike trains are sorted, which is ensured if you use the load_spike_trains_from_txt function with the parameter is_sorted=False. If in doubt, use spike_train = np.sort(spike_train) to obtain a correctly sorted spike train.

Furthermore, the spike trains should have auxiliary spikes at the beginning and end of the observation interval. You can ensure this by providing the time_interval in the load_spike_trains_from_txt function, or calling add_auxiliary_spikes for your spike trains. The spike trains must have the same observation interval!

ISI-distance

The following code loads some exemplary spike trains, computes the dissimilarity profile of the ISI-distance of the first two spike trains, and plots it with matplotlib:

import matplotlib.pyplot as plt
import pyspike as spk

spike_trains = spk.load_spike_trains_from_txt("PySpike_testdata.txt",
                                              time_interval=(0, 4000))
isi_profile = spk.isi_profile(spike_trains[0], spike_trains[1])
x, y = isi_profile.get_plottable_data()
plt.plot(x, y, '--k')
print("ISI distance: %.8f" % isi_profil.avrg())
plt.show()

The ISI-profile is a piece-wise constant function, and hence the function isi_profile returns an instance of the PieceWiseConstFunc class. As shown above, this class allows you to obtain arrays that can be used to plot the function with plt.plt, but also to compute the time average, which amounts to the final scalar ISI-distance. By default, the time average is computed for the whole PieceWiseConstFunc function. However, it is also possible to obtain the average of a specific interval by providing a pair of floats defining the start and end of the interval. In the above example, the following code computes the ISI-distances obtained from averaging the ISI-profile over four different intervals:

isi1 = isi_profil.avrg(interval=(0, 1000))
isi2 = isi_profil.avrg(interval=(1000, 2000))
isi3 = isi_profil.avrg(interval=[(0, 1000), (2000, 3000)])
isi4 = isi_profil.avrg(interval=[(1000, 2000), (3000, 4000)])

Note, how also multiple intervals can be supplied by giving a list of tuples.

If you are only interested in the scalar ISI-distance and not the profile, you can simply use:

isi_dist = spk.isi_distance(spike_trains[0], spike_trains[1], interval)

where interval is optional, as above, and if omitted the ISI-distance is computed for the complete spike trains.

Furthermore, PySpike provides the average_profile function that can be used to compute the average profile of a list of given PieceWiseConstFunc instances.

isi_profile1 = spk.isi_profile(spike_trains[0], spike_trains[1])
isi_profile2 = spk.isi_profile(spike_trains[0], spike_trains[2])
isi_profile3 = spk.isi_profile(spike_trains[1], spike_trains[2])

avrg_profile = spk.average_profile([isi_profile1, isi_profile2, isi_profile3])
x, y = avrg_profile.get_plottable_data()
plt.plot(x, y, label="Average ISI profile")

Note the difference between the average_profile function, which returns a PieceWiseConstFunc (or PieceWiseLinFunc, see below), and the avrg member function above, that computes the integral over the time profile resulting in a single value. So to obtain overall average ISI-distance of a list of ISI profiles you can first compute the average profile using average_profile and the use

avrg_isi = avrg_profile.avrg()

to obtain the final, scalar average ISI distance of the whole set (see also "Computing multivariate distance" below).

SPIKE-distance

To compute for the spike distance you use the function spike_profile instead of isi_profile above. But the general approach is very similar:

import matplotlib.pyplot as plt
import pyspike as spk

spike_trains = spk.load_spike_trains_from_txt("PySpike_testdata.txt",
                                              time_interval=(0, 4000))
spike_profile = spk.spike_profile(spike_trains[0], spike_trains[1])
x, y = spike_profile.get_plottable_data()
plt.plot(x, y, '--k')
print("SPIKE distance: %.8f" % spike_profil.avrg())
plt.show()

This short example computes and plots the SPIKE-profile of the first two spike trains in the file PySpike_testdata.txt. In contrast to the ISI-profile, a SPIKE-profile is a piece-wise linear function and is therefore represented by a PieceWiseLinFunc object. Just like the PieceWiseConstFunc for the ISI-profile, the PieceWiseLinFunc provides a get_plottable_data member function that returns arrays that can be used directly to plot the function. Furthermore, the avrg member function returns the average of the profile defined as the overall SPIKE distance. As above, you can provide an interval as a pair of floats as well as a sequence of such pairs to avrg to specify the averaging interval if required.

Again, you can use

spike_dist = spk.spike_distance(spike_trains[0], spike_trains[1], interval)

to compute the SPIKE distance directly, if you are not interested in the profile at all. The parameter interval is optional and if neglected the whole spike train is used. Furthmore, you can use the average_profile function to compute an average profile of a list of SPIKE-profiles:

avrg_profile = spk.average_profile([spike_profile1, spike_profile2,
                                    spike_profile3])
x, y = avrg_profile.get_plottable_data()
plt.plot(x, y, label="Average SPIKE profile")

Computing multivariate profiles and distances

To compute the multivariate ISI- or SPIKE-profile of a set of spike trains, you can compute all bivariate profiles separately and then use the average_profile function above. However, PySpike provides convenience functions for that purpose. The following example computes the multivariate ISI- and SPIKE-profile for a list of spike trains:

spike_trains = spk.load_spike_trains_from_txt("PySpike_testdata.txt",
                                              time_interval=(0, 4000))
avrg_isi_profile = spk.isi_profile_multi(spike_trains)
avrg_spike_profile = spk.spike_profile_multi(spike_trains)

Both functions take an optional parameter indices, a list of indices that allows to define the spike trains that should be used for the multivariate profile. As before, if you are only interested in the distance values, and not in the profile, PySpike offers the functions: isi_distance_multi and spike_distance_multi, that return the scalar multivariate ISI- and SPIKE-distance. Both distance functions also accept an interval parameter that can be used to specify the begin and end of the averaging interval as a pair of floats, if neglected the complete interval is used.

Another option to characterize large sets of spike trains are distance matrices. Each entry in the distance matrix represents a bivariate distance of two spike trains. The distance matrix is symmetric and has zero values at the diagonal. The following example computes and plots the ISI- and SPIKE-distance matrix, where for the latter one the averaging is performed only the time interval T=0..1000.

spike_trains = spk.load_spike_trains_from_txt("PySpike_testdata.txt", 4000)

plt.figure()
isi_distance = spk.isi_distance_matrix(spike_trains)
plt.imshow(isi_distance, interpolation='none')
plt.title("ISI-distance")

plt.figure()
spike_distance = spk.spike_distance_matrix(spike_trains, interval=(0,1000))
plt.imshow(spike_distance, interpolation='none')
plt.title("SPIKE-distance")

plt.show()

The work on PySpike was supported by the European Comission through the Marie Curie Initial Training Network Neural Engineering Transformative Technologies (NETT) under the project number 289146.

Python/C Programming:

Mario Mulansky

Scientific Methods:

Thomas Kreuz
Nebojsa D. Bozanic

File	Mode	Size
doc
examples
pyspike
test
.travis.yml	-rw-r--r--	200 bytes
Contributors.txt	-rw-r--r--	294 bytes
License.txt	-rw-r--r--	1.3 KB
Readme.rst	-rw-r--r--	13.9 KB
setup.cfg	-rw-r--r--	41 bytes
setup.py	-rw-r--r--	2.1 KB

https://github.com/mariomulansky/PySpike

Readme.rst