Skip to content

Conversation

@bamford
Copy link
Owner

@bamford bamford commented Apr 5, 2023

Local repo continuation of #1.

sdpenguin and others added 21 commits July 8, 2020 21:32
The data argument is passed to determine the save or stored location of
MNIST npy.
The user will now be prompted to train if testing is run in the
absence of a h5 store.
Restructured the parameters and configuration.
* Config is now supplied via an indepenent object
* Moved functions to utilities + Neatened and added descriptions
* Clarified folder paths
* General tidy up
* Added auxiliary cmd line arguments to help with new Config structure
* Fix equation parsing error by not using str on an Equations object
* Refactor to provide a streamlined API for the NeuronGroup and Synapse classes
* Put Parameters as class variables rather than defined in code
where possible
These are not global parameters and as such should not be included
in the config object.
(Simulation Now Working)
The custom_namespace argument parsing has been fixed.
It was only one line.
* Neatened simulation by moving lots of configuration to the config
object instead.
* Removed the redundant record_arguments function, and instead
set the run parameters for metadata using config.__dict__
Moved configuration parameters from the simulation function to the
Config object.
Some variable names were undefined.
This method abstracts the connection forming process and neatens
the get_initial_weights function.
* Redundant parameters have been removed
* Defaults have been set in the Config object and no longer rely
on conditional logic as heavily
Replace : with _ to make folder names better for Windows.
Changes made to my repo by sdpenguin
@bamford
Copy link
Owner Author

bamford commented Apr 5, 2023

Tried python simulation.py --train --size=25. The sdpenguin version runs about half the speed, generates an order of magnitude more spikes, and has lower performance. There must be some differences in the parameters.

Sensible next steps are probably the following:

  • compare details using h5 output file
  • gradually merge in changes, starting with the most straightforward, to narrow in on issue

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants