flatironinstitute
Accuracy of mountainsort without curate=true, and setting interspike intervals
Hello everyone, I am sorting around 5-6s of data per trial from 256 channels, and I have around 320 trials. The number of neurons detected varies in each trial from around 227 to more than 500 without curate. On using curate, the number of cluster decreases to around 3-4. I am sorting each channel independently. Sometimes, a particular channel shows no spikes at all. Is there any way to adjust parameters like interspike intervals and threshold so that I can optimise between the quality of clusters and number of units detected?
5-6s is very short. I don't think the algorithm would be expected to work very well on such a short dataset.
There is a way to customize the curation, but right now it would involve creating a custom processor in place of the current python script.
Could you tell me the approximate time scale for which the algorithm is expected to work correctly? Also, how 'bad' would the results be? Would atleast the spike detections be accurate? Clustering is not very important to me right now.
On Fri, Jun 8, 2018 at 12:01 AM, Jeremy Magland [email protected] wrote:
5-6s is very short. I don't think the algorithm would be expected to work very well on such a short dataset.
There is a way to customize the curation, but right now it would involve creating a custom processor in place of the current python script.
— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/flatironinstitute/mountainsort/issues/49#issuecomment-395474659, or mute the thread https://github.com/notifications/unsubscribe-auth/AVgTZHcchTwRwkgn8RadwLNtUA4-iKX6ks5t6U5AgaJpZM4UdtKw .
Typically you'd like to have at least several hundred spikes per neuron. So, for example, if the spike rate for a neuron is 1 Hz, then you'd want at least several minutes of recording.