Recall for peak detectionr

recall_for_peak_detector.py

def calculate_peaks_spike_recall(peaks, sorting_ground_truth, tolerance_ms=0.4):
    """
    Calculate the spike recall of the peaks (which are the output of a peak detection method)
    against a ground truth of spike_trains. This function is used to test the quality of a peak detection method
    in the context of a specific sorting. 
    
    Recall close to 1 means that all the spike in the spike_train are present in a peak whereas recall
    close to 0 means that no spike in the spike_train are present in a peak. 
    
    More technically, this calculates the number of True positives divided by 
    the total number of positive examples (i.e. the number of spikes) 
    within a given tolerance.
    
    The tolerance is given in milliseconds and means that the detected peak 
    time should be less than the tolerance away from the real spike time.
    
    """    

    sample_indices = peaks["sample_index"]
    spike_trains = sorting_ground_truth.get_all_spike_trains()[0][0]
    sampling_freuency = sorting_ground_truth.get_sampling_frequency()
    tolerance_number_samples = tolerance_ms * sampling_freuency / 1_000.0
    are_spikes_close_any_peaks = np.any(np.abs(sample_indices - spike_trains[:, np.newaxis]) < tolerance_number_samples, axis=1)

    return are_spikes_close_any_peaks.mean()



# Test with this
peaks = peaks_by_channel_np
method_recall = calculate_peaks_spike_recall(peaks, sorting_ground_truth)
pytest.approx(method_recall, 1.0)