Source code for allensdk.brain_observatory.behavior.event_detection

import numpy as np
from scipy import stats


def filter_events_array(arr: np.ndarray, scale: float = 2,
                        n_time_steps: int = 20) -> np.ndarray:
    """
    Convolve the trace array with a 1d causal half-gaussian filter
    to smooth it for visualization

    Uses a halfnorm distribution as weights to the filter

    Modified from initial implementation by Nick Ponvert

    Parameters
    ----------
    arr: np.ndarray
        Trace matrix of dimension n traces x n frames
    scale: float
        std deviation of halfnorm distribution in units of timesteps
    n_time_steps: int
        number of time steps to use for the convolution operation

    Returns
    ----------
    np.ndarray:
        Output of the convolution operation
    """
    if len(arr.shape) == 1:
        raise ValueError('Expected a 2d array but received a 1d array')

    if n_time_steps < 1:
        raise ValueError(f'n_time_steps must be a minimum of 1 but received '
                         f'{n_time_steps}')

    filt = stats.halfnorm(loc=0, scale=scale).pdf(np.arange(n_time_steps))
    filt = filt / np.sum(filt)  # normalize filter
    filtered_arr = np.zeros(arr.shape)
    for i, trace in enumerate(arr):
        filtered_arr[i] = np.convolve(arr[i], filt)[:len(arr[i])]
    return filtered_arr