Closed bucklesm closed 9 months ago
Thanks for checking out the repo @bucklesm!
We're finalizing cleanup now in the dev
branch before moving everything over to main
, there are a ton of updates that we're going to push for a polished v1.0.
Thanks @jasonlyik! I have been playing with the NHP task, in both the main
and dev
branches. I've found it tricky to parse what the input data is as well as some of the processing that seems to happen within the ANN (and how you choose the dividing line for "pre-processing"). I'm looking forward to the v1.0. If there's a document describing the data flow (especially for the NHP benchmark) a little in the meantime, I'd love to take a look.
@bucklesm For the regression tasks NHP and MackeyGlass, we use a slightly different data format where each timestep is considered a different sample, and the prediction is the next timestep data. So the data tensor ends up as [B, T, F]
, B = number of samples (timesteps) to predict
, T = 1 timestep per sample
, F = 96 or 192 spike features
. We can add more detail to the tutorials with the full version up as well!
Just starting with the repo and testing the examples. Following issues:
from neurobench.datasets import MSWC_FSCIL
doesn’t work/exist in datasets.from neurobench.datasets import Prophesee1MP
doesn’t exist in the datasets folder.