Open katie-marquand opened 1 year ago
Sorry for taking so long. I'm afraid I don't have an immediate fix for this but in the future we should be able to implement something in navis-fastcore that will solve this.
Another quick update: I had a crack at this but implemented in rust as part of an experiment in fastcore-rs.
With that I can calculate the synapse flow (which is what failed previously) in ~30s for the hemibrain APL:
(Brighter = higher synapse flow)
Making an axon-dendrite split based off that currently still needs a bit of tinkering to get to work but if I cut some corners, we get something that doesn't look half bad:
(dark blue = dendrites; cyan = axon)
I always assumed the APL simply goes calyx (CA) -> peduncle -> lobes but if that skeleton is topologically correct, then the CA + peduncle are actually isolated from the lobes and it's actually CA + peduncle -> lobes.
If the neuron has too many synapses the geodesic distance matrix can't be initialised because it takes too much memory.