Utils

spykes.utils.circ_corr(alpha1, alpha2)

Helper function to compute the circular correlation.

spykes.utils.get_sort_indices(data, by=None, order='descend')

Helper function to calculate sorting indices given sorting condition.

Parameters:

• data (2-D numpy array) – Array with shape (n_neurons, n_bins).
• by (str or list) – If rate, sort by firing rate. If latency, sort by peak latency. If a list or array is provided, it must correspond to integer indices to be used as sorting indices. If no sort order is provided, the data is returned as-is.
• order (str) – Direction to sort in (either descend or ascend).

Returns:

The sort indices as a Numpy array, with one index per element in data (i.e. data[sort_idxs] gives the sorted data).

Return type:

list

spykes.utils.grad_slow_exp(z, eta)

Computes the gradient of a slowly rising exponential function.

This is defined as:

\[\begin{split}\nabla q = \begin{cases} \exp(eta) & \text{if } z > eta \\ \exp(z) & \text{if } z \leq eta \end{cases}\end{split}\]

Parameters:

• z (array) – The dependent variable, before calling the slow_exp() function.
• eta (float) – The threshold parameter used in the original slow_exp() call.

Returns:

The gradient with respect to z of the output of slow_exp().

Return type:

array

spykes.utils.log_likelihood(y, yhat)

Helper function to compute the log likelihood.

spykes.utils.set_matplotlib_defaults(plt=None)

Sets publication quality defaults for matplotlib.

Parameters:plt (matplotlib.pyplot instance) – The plt instance.

spykes.utils.slow_exp(z, eta)

Applies a slowly rising exponential function to some data.

This function defines a slowly rising exponential that linearizes above the threshold parameter eta. Mathematically, this is defined as:

\[\begin{split}q = \begin{cases} (z + 1 - eta) * \exp(eta) & \text{if } z > eta \\ \exp(eta) & \text{if } z \leq eta \end{cases}\end{split}\]

The gradient of this function is defined in grad_slow_exp().

Parameters:

• z (array) – The data to apply the slow_exp() function to.
• eta (float) – The threshold parameter.

Returns:

The resulting slow exponential, with the same shape as z.

Return type:

array

spykes.utils.train_test_split(*datasets, **split)

Splits test data into training and testing data.

This is a replacement for the Scikit Learn version of the function (which is being deprecated).

Parameters:

• datasets (list of Numpy arrays) – The datasets as Numpy arrays, where the first dimension is the batch dimension.
• n (int) – Number of test samples to split off (only n or percent may be specified).
• percent (int) – Percentange of test samples to split off.

Returns:

If only one dataset is provided, this method returns a tuple of training and testing data; otherwise, it returns a list of such tuples.

Return type:

tuple of train / test data, or list of tuples