teili.stimuli package

Submodules

teili.stimuli.testbench module

This class holds different pre-defined testbench stimuli.

The idea is to test certain aspects of you network with common stimuli.

Example

>>> import numpy as np
>>> from brian2 import us, ms
>>> from pyqtgraph.Qt import QtCore, QtGui
>>> import pyqtgraph as pg
>>> from teili.stimuli.testbench import OCTA_Testbench
>>> from teili.tools.plotter2d import Plotter2d

>>> app = QtGui.QApplication.instance()
>>> if app is None:
        app = QtGui.QApplication(sys.argv)
>>> else:
        print('QApplication instance already exists: %s' % str(app))

>>> testbench = OCTA_Testbench()
>>> testbench.rotating_bar(length=10, nrows=10, direction='ccw', ts_offset=3,
                       angle_step=10, noise_probability=0.2, repetitions=90, debug=False)

In order to visualize it:

>>> event_monitor = Plotter2d.loaddvs(testbench.events)
>>> imv1 = event_monitor.plot3d_on_off(plot_dt=10*ms, filtersize=15*ms)

>>> win = pg.GraphicsWindow(title="DVS Spikes")
>>> gridlayout = QtGui.QGridLayout(win)
>>> gridlayout.addWidget(imv1, 1, 1)
>>> win.resize(1500, 1000)
>>> win.setLayout(gridlayout)
>>> win.show()
>>> win.setWindowTitle('DVS plot')
>>> imv1.play(10)

>>> app.exec_()

Todo

  • As soon as visualizer class is updated, change imports!

class teili.stimuli.testbench.OCTA_Testbench(DVS_SHAPE=(240, 180))[source]

Bases: object

This class holds all relevant stimuli to test modules provided with the Online Clustering of Temporal Activity (OCTA) framework.

angles

List of angles of orientation.

Type

numpy.ndarray

DVS_SHAPE

Input shape of the simulated DVS/DAVIS vision sensor.

Type

TYPE

end

End pixel location of the line.

Type

TYPE

events

Attribute storing events of testbench stimulus.

Type

TYPE

indices

Attribute storing neuron index of testbench stimulus.

Type

TYPE

line

Stimulus of the testbench which is used to either generate an interactive plot to record stimulus with a DVS/DAVIS camera or coordinates are used to generate a SpikeGenerator.

Type

TYPE

start

Start pixel location of the line.

Type

TYPE

times

Attribute storing spike times of testbench stimulus.

Type

list

aedat2events(rec, camera='DVS128')[source]

Wrapper function of the original aedat2numpy function in teili.tools.converter.

This function will save events for later usage and will directly return them if no SpikeGeneratorGroup is needed.

Parameters

  • rec (str) – Path to stored .aedat file.

  • camera (str, optional) – Can either be string (‘DAVIS240’) or int 240, which specifies the larger of the 2 pixel dimension to unravel the coordinates into indices.

Returns

4D numpy array with #events entries. Array is organized as x, y, ts, pol. See aedat2numpy for more details.

Return type

events (np.ndarray)

ball(rec_path)[source]

This function loads a simple recording of a ball moving in a small arena. The idea is to test the Online Clustering and Prediction module of OCTAPUS. The aim is to learn spatio-temporal features based on the ball’s trajectory and learn to predict its movement.

Parameters

rec_path (str, required) – Path to recording.

Returns

A SpikeGenerator which has index (i) and spiketimes (t) as attributes

Return type

SpikeGeneratorGroup (brian2.obj)

Raises

UserWarning – If no filename is given but aedat reacording should be loaded

dda_round(x)[source]

Simple round funcion.

Parameters

x (float) – Value to be rounded.

Returns

Ceiled value of x.

Return type

(int)

infinity(cAngle)[source]

Given an angle cAngle this function returns the current position on an infinity trajectory.

Parameters

cAngle (float) – current angle in rad which determines position on infinity trajectory.

Returns

Postion in x, y coordinates.

Return type

position (tuple)

rotating_bar(length=10, nrows=10, ncols=None, direction='ccw', ts_offset=10, angle_step=10, artifical_stimulus=True, rec_path=None, save_path=None, noise_probability=None, repetitions=1, debug=False)[source]

This function returns a single SpikeGeneratorGroup (Brian object).

The purpose of this function is to provide a simple test stimulus. A bar is rotating in the center. The goal is to learn necessary spatio-temporal features of the moving bar and be able to make predictions about where the bar will move.

Parameters

  • length (int) – Length of the bar in pixel.

  • nrows (int, optional) – X-Axis size of the pixel array.

  • ncols (int, optional) – Y-Axis size of the pixel array.

  • orientation (str) – Orientation of the bar. Can either be ‘vertical’ or ‘horizontal’.

  • ts_offset (int) – time between two pixel location.

  • angle_step (int, optional) – Angular velocity. Sets step width in np.arrange.

  • artifical_stimulus (bool, optional) – Flag if stimulus should be created or loaded from aedat file.

  • rec_path (str, optional) – Path/to/stored/location/of/recorded/stimulus.aedat.

  • save_path (str, optional) – Path to store generated events.

  • noise_probability (float, optional) – Probability of noise events between 0 and 1.

  • repetitions (int, optional) – Number of revolutions of the rotating bar.

  • debug (bool, optional) – Flag to print more detailed output of testbench.

Returns

Brian2 objects which holds the spike times as well

as the respective neuron indices

Return type

SpikeGenerator obj

Raises

UserWarning – If no filename is given but aedat recording should be loaded

rotating_bar_infinity(length=10, nrows=64, ncols=None, orthogonal=False, shift=32, ts_offset=10, artifical_stimulus=True, rec_path=None, return_events=False)[source]

This function will either load recorded DAVIS/DVS recordings or generate artificial events of a bar moving on an infinity trajectory with fixed orientation, i.e. no super-imposed rotation. In both cases, the events are provided to a SpikeGeneratorGroup which is returned.

Parameters

  • length (int, optional) – Length of the bar in pixel.

  • nrows (int, optional) – X-Axis size of the pixel array.

  • ncols (int, optional) – Y-Axis size of the pixel array.

  • orthogonal (bool, optional) – Flag which determines if bar is kept always orthogonal to trajectory, if it kept aligned with the trajectory or if it returns in a “chaotic” way.

  • shift (int, optional) – Offset in x where the stimulus will start.

  • ts_offset (int, optional) – Time in ms between consecutive pixels (stimulus velocity).

  • artifical_stimulus (bool, optional) – Flag if stimulus should be created or loaded from aedat file.

  • rec_path (str, optional) – Path/to/stored/location/of/recorded/stimulus.aedat.

  • return_events (bool, optional) – Flag to return events instead of SpikeGenerator.

Returns

A SpikeGenerator which has index (i) and spiketimes (t) as attributes. events (numpy.ndarray, optional): If return_events is set, events will be returned.

Return type

SpikeGeneratorGroup (brian2.obj)

Raises

UserWarning – If no filename is given but aedat recording should be loaded.

translating_bar_infinity(length=10, nrows=64, ncols=None, orientation='vertical', shift=32, ts_offset=10, artifical_stimulus=True, rec_path=None, return_events=False)[source]

This function will either load recorded DAVIS/DVS recordings or generate artificial events of a bar moving on an infinity trajectory with fixed orientation, i.e. no super-imposed rotation. In both cases, the events are provided to a SpikeGeneratorGroup which is returned.

Parameters

  • length (int, optional) – length of the bar in pixel.

  • nrows (int, optional) – X-Axis size of the pixel array.

  • ncols (int, optional) – Y-Axis size of the pixel array.

  • orientation (str, optional) – lag which determines if bar is orientated vertically or horizontally.

  • shift (int, optional) – offset in x where the stimulus will start.

  • ts_offset (int, optional) – Time in ms between consecutive pixels (stimulus velocity).

  • artifical_stimulus (bool, optional) – Flag if stimulus should be created or loaded from aedat file.

  • rec_path (str, optional) – Path/to/stored/location/of/recorded/stimulus.aedat.

  • return_events (bool, optional) – Flag to return events instead of SpikeGenerator.

Returns

A SpikeGenerator which has index (i) and spiketimes (t) as attributes. events (numpy.ndarray, optional): If return_events is set, events will be returned.

Return type

SpikeGeneratorGroup (brian2.obj)

Raises

UserWarning – If no filename is given but aedat recording should be loaded.

class teili.stimuli.testbench.STDGM_Testbench(N=1, stimulus_length=1200)[source]

Bases: object

This class provides a stimulus to test your spike-timing dependent gain modulation algorithm.

stimuli(isi)[source]

Stimulus gneration for STDGM protocols.

This function returns two brian2 objects. Both are Spikegeneratorgroups which hold a single index each and varying spike times. The protocol follows homoeostasis, weak LTP, weak LTD, strong LTP, strong LTD, homoeostasis.

Parameters

isi (int, optional) – Interspike Interval. How many spikes per stimulus phase.

Returns

Brian2 objects which hold the spiketimes and

the respective neuron indices.

Return type

SpikeGeneratorGroup (brian2.obj

class teili.stimuli.testbench.STDP_Testbench(N=1, stimulus_length=1200)[source]

Bases: object

This class provides a stimulus to test your spike-timing dependent plasticity algorithm.

N

Size of the pre and post neuronal population.

Type

int

stimulus_length

Length of stimuli in ms.

Type

int

stimuli(isi=10)[source]

Stimulus gneration for STDP protocols.

This function returns two brian2 objects. Both are Spikegeneratorgroups which hold a single index each and varying spike times. The protocol follows homoeostasis, weak LTP, weak LTD, strong LTP, strong LTD, homoeostasis.

Parameters

isi (int, optional) – Interspike Interval. How many spikes per stimulus phase.

Returns

Brian2 objects which hold the spiketimes and

the respective neuron indices.

Return type

SpikeGeneratorGroup (brian2.obj

class teili.stimuli.testbench.SequenceTestbench(n_channels, n_items, cycle_length, noise_probability=None, rate=None)[source]

Bases: object

This class provides a simple poisson encoded sequence testbench. This class returns neuron indices and spike times, which can used to instantiate a SpikeGeneratorGroup in brian2.

add_noise()[source]

This function adds noise spike given the noise_probability.

create_poisson_items()[source]

This function creates the Poisson distributed items with the specificed rate.

stimuli()[source]

This function creates the stimuli and returns neuron indices and spike times.

class teili.stimuli.testbench.WTA_Testbench[source]

Bases: object

Collection of functions to test the computational properties of the WTA building_block.

indices

Array with neuron indices.

Type

numpy.ndarray

noise_input

PoissonGroup which provides noise events.

Type

brian2.PoissonGroup

times

Array with neuron spike times.

Type

numpy.ndarray

background_noise(num_neurons=10, rate=10)[source]

Provides background noise as Poisson spike trains

Parameters

  • num_neurons (int, optional) – 1D size of WTA population.

  • rate (int, optional) – Spike frequency f Poisson noise process.

stimuli(num_neurons=16, dimensions=2, start_time=10, end_time=500, isi=2)[source]

This function provides simple test stimuli to test the selection mechanism of a WTA population.

Parameters

  • num_neurons (int, optional) – 1D size of WTA population.

  • dimensions (int, optional) – Dimension of WTA. Can either be 1 or 2

  • start_time (int, optional) – Start time when stimulus should start.

  • end_time (int, optional) – End time when stimulus should stop.

  • isi (int, optional) – Inter-spike between spike times.

Raises

NotImplementedError – If dimension is not 1 or 2 this error is raised

Module contents