Hi all,
Thanks for the excellent package! I am interested in calculating ISI distances between the same neuron's spike trains across trials of differing lengths. In isi_profile_bi(), the interval is assumed to be spike_train1.t_start to spike_train1.t_end (I assume similar for the other metrics). reconcile_spike_trains() takes the earliest start time and the latest end time. If I understand the implementation correctly, both of these are highly problematic because times when spikes can be occurring in spike_train1 may be compared to times when it's impossible to know whether spike_train2 had spikes, but it's assumed there were zero spikes. The basic fix is to only allow calculating pairwise distance between spike trains when both spike trains were observed (i.e., taking max of start times and min of end times).
There are some other assumptions that are violated when taking the average across the multivariate case for pairwise distances of trials of differing lengths (i.e., some time bins are observed more frequently, and are overrepresented in the average) but a word of warning in the documentation might suffice for that more complex case.
Best,
Charlie
Hi all,
Thanks for the excellent package! I am interested in calculating ISI distances between the same neuron's spike trains across trials of differing lengths. In
isi_profile_bi(), the interval is assumed to be spike_train1.t_start to spike_train1.t_end (I assume similar for the other metrics).reconcile_spike_trains()takes the earliest start time and the latest end time. If I understand the implementation correctly, both of these are highly problematic because times when spikes can be occurring in spike_train1 may be compared to times when it's impossible to know whether spike_train2 had spikes, but it's assumed there were zero spikes. The basic fix is to only allow calculating pairwise distance between spike trains when both spike trains were observed (i.e., taking max of start times and min of end times).There are some other assumptions that are violated when taking the average across the multivariate case for pairwise distances of trials of differing lengths (i.e., some time bins are observed more frequently, and are overrepresented in the average) but a word of warning in the documentation might suffice for that more complex case.
Best,
Charlie