Closed russelljjarvis closed 6 years ago
@russelljjarvis I added lots of comments in the "Files Changed" section that you should look over and respond to (or address) so we can get closer to merging this PR.
The code seems slower with NEURONMemoryBackend commented out. I wonder if it's okay if I move this to a seperate file, and unit test + time, with and without this backend in order to objectively quantify the time difference.
Are you using my caching code (i.e. does your Backend parent class have a local_run that checks the cache first? If so, it sounds like maybe your NEURONMemoryBackend is doing caching in scenarios that mine isn't. Maybe you can just setup a unit test that invokes the Neuron model a few times, and prints the duration of each run, and the expected duration based on whether caching should be occurring or not. Then the same thing can be run with the generic backend and we can compare.
Make distance between branches