early_vision_toolbox.tuning

functions for measuring tuning of model neurons as in real experiments

class early_vision_toolbox.tuning.LinearSquareNeuronBank(w)[source]

Bases: early_vision_toolbox.interface.NeuronBank

NeuronBank with linear response computing methods and square RF size

Attributes

n_neuron number of neurons
rf_size size of RF

Methods

predict(imgs[, neuron_idx]) get neuron response to images
n_neuron

number of neurons

predict(imgs, neuron_idx=None)[source]

get neuron response to images

Parameters:imgs
rf_size

size of RF

early_vision_toolbox.tuning.find_optimal_paras_rf(w, freqvalues=None, orvalues=None, phasevalues=None, legacy=False)[source]
Parameters:

w : numpy.ndarray or a NeuronBank object

if a 2d array, then each row representing a flattened filter, in row major order. and response will be evaluated in a linear fasion. Otherwise, it must be a NeuronBank object.

freqvalues, orvalues, phasevalues : array_like or None, optional

(implicitly flattened) 1d array of frequency, orientation, and phase values to test. By default, they have values as used in the demo program (figures.m) of the original code (see Notes):

%set number of different values for the grating parameters used in
%computing tuning curves and optimal parameters
freqno=50; %how many frequencies
orno=50; %how many orientations
phaseno=20; %how many phases
%compute the used values for the orientation angles and frequencies
orvalues=[0:orno-1]/orno*pi;
freqvalues=[0:freqno-1]/freqno*patchsize/2;
phasevalues=[0:phaseno-1]/phaseno*2*pi;

legacy: bool, optional

whether returning exactly the same optx and opty as in the original implementation (which I believe is problematic). Default False also, this will use the original implementation for computing grating.
Returns:

a numpy structured array with 5 columns, each row for a filter. It has the following dtype.

[('optx', np.float64),  #  gravity center of x (column)
 ('opty', np.float64),  #  gravity center of y (ro)
 ('optfreq', np.float64),  #  optimal frequency
 ('optor', np.float64),    #  optimal orientation.
 ('optphase', np.float64)] #  optimal phase

Notes

The linear part of this function is basically a port of findoptimalparas.m from code for [R1].

References

[R1](1, 2) Hyvärinen, A., Hurri, J., & Hoyer, P. O. (2009). Natural Image Statistics: A Probabilistic Approach to Early Computational Vision. (1st ed., Vol. 39). Springer Publishing Company, Incorporated. Retrieved from http://www.naturalimagestatistics.net/ http://dx.doi.org/10.1007/978-1-84882-491-1
early_vision_toolbox.tuning.freq_tuning_curve(w, freqvalues_test=None, **kwargs)[source]

find the frequency tuning of a bunch of neurons.

Parameters:

w

freqvalues_test : array_like

kwargs
Returns:

a 2d numpy ndarray, each row being the response of neuron to freqvalues_probe
early_vision_toolbox.tuning.ori_tuning_curve(w, orvalues_test=None, **kwargs)[source]

find the orientation tuning of a bunch of neurons.

Parameters:

w

orvalues_test : array_like

kwargs
Returns:

a 2d numpy ndarray, each row being the response of neuron to freqvalues_probe
early_vision_toolbox.tuning.phase_tuning_curve(w, phasevalues_test=None, **kwargs)[source]

find the phase tuning of a bunch of neurons.

Parameters:

w

freqvalues_probe : array_like

kwargs
Returns:

a 2d numpy ndarray, each row being the response of neuron to freqvalues_probe