Distance dependent structural plasticity #3369
Open
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
gtrensch
added
T: Enhancement
|
Looks good. We might reduce some overhead by passing a pointer to the relevant element in the global_position vector along with its dimension directly to the gaussianKernel function instead of creating and passing subvectors on each iteration, but for a small number of neurons it would not be noticeable |
jhnnsnk
suggested changes
Mar 21, 2025
There was a problem hiding this comment.
Thank you very much for this PR! At a first glance, I have the following comments:
- I suggest to improve the interface a bit: Can
nest.EnableStructuralPlasticity()get arguments instead of havingnest.structural_plasticity_gaussian_kernel_sigmaandnest.structural_plasticity_cache_probabilitiesseparately? - Using
structural_plasticity_gaussian_kernel_sigma = -1as default is misleading. How about an argument such asuse_Gaussian_kernel=trueinstead and then allowing only values for sigma that actually make sense (and allowing to set this value only if the Gaussian kernel is enabled)? - How does the performance compare if you connect and simulate without the Gaussian kernel? And with and without cached probabilities?
- Could you please document what caching probabilities actually does?
- I suggest to move the pytests for the spatial version into a separate file in
test_sp. - The test
test_distance_dependent_without_positionsappears twice. - How are autapses and multapses handled during the creation and later during the simulation?
- Can masks be supported (as a maximum distance)?
github-project-automation
bot
moved this from In progress
to PRs pending approval
in Kernel
|
@alpinangu ,@jhnnsnk Thank you so much for your feedback! I'm working on the suggested changes and the performance comparisons. |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Description
This pull request introduces new functionality to handle distance-dependent connection probabilities in the
SPManagerclass. The following methods have been added or modified:New Features
gather_global_positions_and_ids()Collects and verifies global neuron positions and IDs.
Ensures that neurons are spatially distributed, raising an error if positions are undefined while distance dependency is enabled.
gaussian_kernel()Computes a probability based on the Gaussian distance kernel between two neuron positions.
build_probability_list()Constructs a probability list for all neuron pairs based on their spatial positions.
Supports caching for efficient repeated lookups.
get_neuron_pair_index()Maps two neuron IDs to a unique index for accessing pair-specific probabilities in the
probability_list.global_shuffle_spatial()Implements distance-dependent shuffling for pairing neurons based on spatial proximity and probability.
roulette_wheel_selection()Performs a weighted random selection of post-synaptic neurons based on precomputed probabilities.
Modified Methods
enable_structural_plasticity()Integrates calls to
gather_global_positions_and_ids()andbuild_problist()when distance dependency is enabled.Distance-Dependent Features
Enable Distance Dependency
Distance dependency can be enabled by setting
structural_plasticity_gaussian_kernel_sigmato a positive value. This parameter controls the width of the Gaussian kernel used for connection probabilities. Higher values result in reduced sensitivity to distance, making distant neurons more likely to connect.Caching Connection Probabilities
If
structural_plasticity_cache_probabilitiesis set totrue, connection probabilities will be precomputed and cached. This can improve performance, particularly in long-running simulations.Testing
The following tests have been added to validate the new functionality:
C++ Tests:
test_gaussian_kernelTests Gaussian kernel computations for various distances and sigma values.
test_get_neuron_pair_indexConfirms correct mapping of neuron pairs to unique indices.
test_global_shuffle_spatialValidates spatially aware shuffling logic.
Ensures connections favor spatial proximity and prevent self-connections.
Python Tests:
test_distance_dependent_without_positions()Ensures an error is raised when distance dependency is enabled, but positions are not provided.
test_gaussian_kernel()Tests probability computation using Gaussian distance dependency.
test_probability_list_building()Verifies the correct initialization and population of the probability list.
test_roulette_wheel_selection()Confirms that weighted random selection behaves as expected.
test_structural_plasticity_with_positions()Ensures structural plasticity works correctly when neuron positions are defined.
test_synapse_creation_distance_dependent()Validates distance-dependent synapse formation using position data.
Related Issue:
Incorporate Distance-Dependent Connection Probability in Structural Plasticity Module #3209