Andrew Sedler
Andrew Sedler
Hi! Does this happen immediately or after a few training steps? I think it's likely that the learning rate is too high. I would try running just a single model...
This lr_init parameter controls the initial learning rate, which will be reduced over the course of training. I’d recommend trying a few different values (1e-3, 3e-4, 1e-4, 3e-5, etc.) and...
Hmmm… what are your batch size, sequence length, and number of neurons? It looks like sequence length may be close to 1k. 100-300 steps is more typical. You’d probably find...
Hmm, could you upload your files to Google Drive and send me a link to [email protected]?
Hey @Riverside-ms just checking in-- were you able to resolve this?
Hey @rohitrangwani, can you first try setting up a [single run script](https://github.com/arsedler9/lfads-torch/blob/main/scripts/run_single.py) using your configs and see if that runs without errors?
Hey @rohitrangwani, sorry for the very long delay. Is this still an issue for you?
Also very interested in this! Has any progress been made?
Hey @wwrlearner, the `lfads-torch` paper didn't have a supplement or synthetic data, but maybe you're referring to [this repo](https://github.com/snel-repo/expressive-latent-dynamics-paper)? You can find some synthetic datasets there.
Hi @wwrlearner, this part of the script is trying to plot `RasterPlot.n_samples` examples of spikes, inferred rates, and inferred inputs. In this case, `RasterPlot.n_samples=3` (the default) and it looks like...