Add option to order Neuropixels channels by depth #513
Conversation
- Changed ChannelHelper method name to be more specific, that it only applies to Neuropixels channels - Move private RawToChannel array to where it needs to be - Assign the channelOrder array a single time before creating the observable, in one place, and it can never be null
|
The main challenge here is mapping between the channel index (i.e., The initial idea (issue comment) was to use a dictionary with the channel as the key and its row/column as the value. I explored alternatives but couldn’t avoid this without pushing the complexity to O(N²), since we need to retrieve the row/column from a given channel index to remap for spatial ordering. We could do this in one pass, but it would require four nested loops (row/col iteration + searching for the matching channel), which might not be worth the tradeoff. The current approach uses O(N) time for this mapping step and O(N log N) overall due to sorting (ref). The dictionary increases space complexity, but it avoids the nested search. I also removed a redundant remapping operation and verified the algorithm on a 2×3 array by hand, which matched the expected behavior (testing with the |
2 participants
This PR adds the ability for data nodes to
OrderByDepth, so that instead of following the conventional channel ordering (0 -> 383), the channels are instead ordered by their physical location along the shank. Therefore, if channel 0 corresponds to electrode 384 (Bank B), but all other channels are on Bank A (1 -> 383), then the channels will be ordered so that channel 0 is electrode 1 (normally channel 1), channel 1 is electrode 2 (normally channel 2), and so on until the very last channel is electrode 384 (normally channel 0).This option is added as a boolean to all data nodes, and is
falseby default, so that if the user loads an old workflow but doesn't change anything, their channel map remains unchanged.The algorithm logic is first proposed in the issue itself, and has been updated to reflect the current logic in the code.
Note: Currently this is only tested in the
NeuropixelsV1edevices, so that we can confirm the logic/structure before applying to all other devices. The Neuropixels devices are easily implemented in the same manner as theNeuropixelsV1e, as they all share a common structure.Fixes #440