neurotools.spikes.convolutional module

Routines for spike-triggered averagind and coherence analysis between spike trains and continuous signals.

Apologies, most of these functions are not yet documented.

neurotools.spikes.convolutional.ccor(i, j, spikes)

Get zero-lag cross-correlation between spike trains i and j.

Parameters:

• i (int) – first neuron index

• j (int) – second neuron index

• spikes (NTRIALS×NNEURONS×NSAMPLES np.array) – Spiking population data

Returns:

x

Return type:

float

neurotools.spikes.convolutional.ccm(i, j, k, spikes)

Construct size k-lag cross-correlation matrix between two spike trains i and j.

Parameters:

• i (int) – first neuron index

• j (int) – second neuron index

• k (positive int) – Number of cross-correlation lags to include

• spikes (NTRIALS×NNEURONS×NSAMPLES np.array) – Spiking population data

Return type:

result

neurotools.spikes.convolutional.blockccm(k, spikes)

Generate covariance matrix for linear least squares. It is a block matrix of all pairwise cross-correlation

Parameters:

• k (positive int) – Number of cross-correlation lags to include

• spikes (NTRIALS×NNEURONS×NSAMPLES np.array) – Spiking population data

Return type:

result

neurotools.spikes.convolutional.sta(i, spikes, lfp)

Construct Spike-Triggered Average (STA) of sigal lfp

Parameters:

• i (int) – Neuron index

• spikes (NTRIALS×NNEURONS×NSAMPLES np.array) – Spiking population data

• lfp (1D np.array)

Return type:

result

neurotools.spikes.convolutional.blocksta(k, spikes, lfp)

Block spike-triggered average vector for time-domain least squares filter NTrials,NNeurons,NSamples = np.shape(spikes)

Parameters:

• k

• spikes

• lfp

Return type:

result

neurotools.spikes.convolutional.reconstruct(k, B, spikes)

Reconstructs LFP from spikes NTrials,NNeurons,NSamples = np.shape(spikes)

Parameters:

• k

• B

• spikes

neurotools.spikes.convolutional.cspect(i, j, spikes)

Get cross-spectral density as FT of cross-correlation

Parameters:

• i

• j

• spikes

Return type:

result

neurotools.spikes.convolutional.cspectm(spikes)

Get all pairs cross spectral matrix NTrials,NNeurons,NSamples = np.shape(spikes)

TODO: This is doing much more work than is needed, and needs to be re-written to operate in the frequency domain.

Parameters:

spikes

Return type:

result

neurotools.spikes.convolutional.spike_lfp_filters(spikes, lfp)

Cross-spectral densities between spikes and LFP NTrials,NNeurons,NSamples = np.shape(spikes)

Parameters:

• spikes

• lfp

Return type:

result

neurotools.spikes.convolutional.spectreconstruct(k, B, spikes=None, fftspikes=None)

Reconstructs LFP from spikes using cross-spectral matrix. Can optionally pass the fts if they are already available NTrials,NNeurons,NSamples = np.shape(spikes)

Parameters:

• k

• B

• spikes (default None)

• fftspikes (default None)

Return type:

result

neurotools.spikes.convolutional.create_spectral_model(spikes, lfp, shrinkage=0)

Parameters:

• spikes

• lfp

• shrinkage (non-negative float; default 0) – L2 regularization strenth

Return type:

result

neurotools.spikes.convolutional.construct_lowpass_operator(fb, k, Fs=1000.0)

TODO: documentation

Constructs a low-pass regularization operator Get the impulse response of a low-pass filter first Then copy it into the matrix. This really only makes sense in the time domain.

Parameters:

• fb

• k

• Fs (positive float; default 1000.) – Sampling rate

Return type:

result

neurotools.spikes.convolutional.autocorrelation_bayes(s, D=200, prior_var=None)

Computes autocorrelation of signal s over time lags D, applying a Gaussian prior of mean zero and variance prior_var. If prior_var is None, then the length of signal s is used.

Parameters:

• s – sequence of values

• D – number of lags to compute, default is 200

• prior_var – positive scalar or None, default is None

Returns:

autocorrelation over lags D, with zero-lag variance

Return type:

xc