Skip to content

Spinncer weight scale#1386

Draft
andrewgait wants to merge 307 commits intomasterfrom
spinncer_weight_scale
Draft

Spinncer weight scale#1386
andrewgait wants to merge 307 commits intomasterfrom
spinncer_weight_scale

Conversation

@andrewgait
Copy link
Contributor

@andrewgait andrewgait commented Sep 25, 2023 •
edited
Loading

Combines both the cerebellum updates #1374 and the weight scaling updates #852 into a single branch for testing on the cerebellum network in the SpiNNCer repository; also attempts to better incorporate some stochastic rounding.

  • The stdp_curve_cond example in PyNN8Examples/examples has required some fine-tuning of the parameters and I'm still not convinced it's good enough to pass as similar enough to the results on other branches
  • The microcircuit model no longer runs in real time on this branch; it requires a time scale factor of 2. This could be investigated further.
  • I am still investigating the effects of this branch on the full-scale cerebellum network which can be run from the SpiNNCer repository (see the README file for details on how to run it). Preliminary results suggest that it can be run successfully at a time scale factor of 1000, but not a time scale factor of 100.

Merge with
SpiNNakerManchester/SpiNNFrontEndCommon#1118
SpiNNakerManchester/PyNNExamples#110
SpiNNakerManchester/sPyNNakerNewModelTemplate#107

rowleya and others added 30 commits May 19, 2020 16:27
@rowleya
Return the provenance
b944223
@rowleya
Fix unit tests
1284951
@rowleya
Need to pass the parameters!
ad128f8
@andrewgait
Merge remote-tracking branch 'origin/master' into weight_scale
25cd5ca 
Conflicts:
	neural_modelling/src/neuron/c_main.c
	neural_modelling/src/neuron/plasticity/stdp/weight_dependence/weight.h
	neural_modelling/src/neuron/plasticity/stdp/weight_dependence/weight_multiplicative_impl.c
	neural_modelling/src/neuron/plasticity/stdp/weight_dependence/weight_multiplicative_impl.h
	neural_modelling/src/neuron/synapses.c
	neural_modelling/src/neuron/synapses.h
	spynnaker/pyNN/models/neural_projections/connectors/abstract_generate_connector_on_machine.py
	spynnaker/pyNN/models/neural_projections/connectors/from_list_connector.py
	spynnaker/pyNN/models/neuron/abstract_population_vertex.py
	spynnaker/pyNN/models/neuron/plasticity/stdp/common/plasticity_helpers.py
	spynnaker/pyNN/models/neuron/plasticity/stdp/timing_dependence/timing_dependence_pfister_spike_triplet.py
	spynnaker/pyNN/models/neuron/plasticity/stdp/timing_dependence/timing_dependence_spike_nearest_pair.py
	spynnaker/pyNN/models/neuron/plasticity/stdp/timing_dependence/timing_dependence_spike_pair.py
	spynnaker/pyNN/models/neuron/plasticity/stdp/timing_dependence/timing_dependence_vogels_2011.py
	spynnaker/pyNN/models/neuron/synaptic_manager.py
	unittests/model_tests/neuron/test_synaptic_manager.py
@andrewgait
flake8 line lengths, unused imports
d4132ff
@andrewgait
Updates to fix integration tests, particularly for kernel connectors
20d210a 
- also fixes for structural plasticity
@andrewgait
Deal with array length of 1
be89efd
@andrewgait
flake8 whitespaces
979891a
@andrewgait
Check representable weights properly when weight_scale isn't 1
459ff56
@andrewgait
Also need to use (float) gcd in STDP cases for better representation
780d513
@pabogdan
tagged IMPLICIT WEIGHT SCALING with a comment
3802b28
@pabogdan
round to nearest wherever I can
7f11fbf
@pabogdan
exponential time constant is now an unsigned long fract
c06ecdd
@pabogdan
potentially fixed v recording
2aa4f94
@pabogdan
correct recording of v
129a45e
@andrewgait
Merge remote-tracking branch 'origin/master' into weight_scale
8a4991d
@andrewgait
Again Python 2 appears to be doing something weird
01a55b4
@andrewgait
Merge remote-tracking branch 'origin/master' into weight_scale
9851271 
Conflicts:
	neural_modelling/src/neuron/plasticity/stdp/weight_dependence/weight_additive_one_term_impl.c
	neural_modelling/src/neuron/plasticity/stdp/weight_dependence/weight_additive_two_term_impl.c
	neural_modelling/src/neuron/plasticity/synapse_dynamics_static_impl.c

(Add a missing UNUSED in weight_multiplicative_impl.c)
@andrewgait
Got the merge slightly wrong...
b3da07b
@andrewgait
Wrong argument again...
16bad5d
@andrewgait
Merge remote-tracking branch 'origin/master' into weight_scale
611148f 
Conflicts:
	neural_modelling/src/neuron/plasticity/stdp/synapse_dynamics_stdp_mad_impl.c
@andrewgait
Merge remote-tracking branch 'origin/master' into weight_scale
fb027ec
@andrewgait
Merge remote-tracking branch 'origin/master' into weight_scale
5db07d9 
Conflicts:
	neural_modelling/src/neuron/c_main.c
	neural_modelling/src/neuron/synapses.c
	neural_modelling/src/neuron/synapses.h
	spynnaker/pyNN/external_devices_models/external_device_lif_control.py
	spynnaker/pyNN/external_devices_models/external_device_lif_control_vertex.py
	spynnaker/pyNN/models/neuron/abstract_population_vertex.py
	spynnaker/pyNN/models/neuron/abstract_pynn_neuron_model.py
	spynnaker/pyNN/models/neuron/abstract_pynn_neuron_model_standard.py
	spynnaker/pyNN/models/neuron/synaptic_manager.py
	spynnaker/pyNN/spynnaker.cfg
	unittests/mocks.py
	unittests/model_tests/neuron/test_synaptic_manager.py
	unittests/test_populations/test_vertex.py
@andrewgait
Use a better tolerance for float gcd, and give constant a name
baf36a1
@andrewgait
flake8 unused import
4627b95
@andrewgait
Get closest representable weight before doing gcd, edit tolerance
687f791
@andrewgait
Fiddle with tolerance again
b817ead
@andrewgait
Merge remote-tracking branch 'origin/master' into weight_scale
e764529
@andrewgait
Avoid dividing by zero...
9e5ac2a
@andrewgait
Merge remote-tracking branch 'origin/master' into weight_scale
d42a790 
Conflicts:
	neural_modelling/src/neuron/population_table/population_table_binary_search_impl.c
	spynnaker/pyNN/models/neuron/abstract_population_vertex.py
	spynnaker/pyNN/models/neuron/synaptic_manager.py
	unittests/model_tests/neuron/test_synaptic_manager.py
andrewgait added 21 commits June 23, 2023 12:36
@andrewgait
Upgrade actions OS (to use a newer compiler)
9ac6bb8
@andrewgait
Merge remote-tracking branch 'origin/master' into weight_scale
7460fb8
@andrewgait
Merge remote-tracking branch 'origin/master' into spinncer_update
30c0127
@andrewgait
Update license
c2cbfaa
@andrewgait
Merge branch 'weight_scale' into spinncer_weight_scale
0cc50f5
@andrewgait
Merge remote-tracking branch 'origin/weight_scale' into
a54c0a2 
spinncer_weight_scale
@andrewgait
Merge remote-tracking branch 'origin/master' into spinncer_weight_scale
5afe281
@andrewgait
Fix merge
257ce82
@andrewgait
flake8
f381fb6
@andrewgait
Not sure how this license got missed in merge
0161cb0
@andrewgait
Add round-to-nearest multiplies back in neuron components
9d0022f
@andrewgait
Adding rounding here should work but causes UNDEF errors?
69a5ec4
@andrewgait
Add comment about UNDEF
da5a0b3
@andrewgait
Merge remote-tracking branch 'origin/master' into spinncer_weight_scale
a49b959
@andrewgait
Merge remote-tracking branch 'origin/master' into spinncer_weight_scale
1bc1f03
@andrewgait
Use stochastic rounding everywhere, particularly for synapse row
0cb28ac
@andrewgait
Add previous method back in comment
17339cb
@andrewgait
Merge remote-tracking branch 'origin/master' into spinncer_weight_scale
f09928f
@andrewgait
Try this instead
e9a6956
@andrewgait
Use absolute value for min weight change and fix test
6d8ec3a
@andrewgait
Merge remote-tracking branch 'origin/master' into spinncer_weight_scale
08e6e26
@coveralls
Copy link

coveralls commented Sep 26, 2023 •
edited
Loading

Coverage Status

coverage: 60.866% (-0.5%) from 61.322% when pulling f25485a on spinncer_weight_scale into 7f5f333 on master.

@andrewgait andrewgait mentioned this pull request Sep 28, 2023
Open
@andrewgait andrewgait linked an issue Sep 28, 2023 that may be closed by this pull request
Ring buffer sigma and spikes-per-second on a per synapse-type basis #136
Open
@andrewgait andrewgait mentioned this pull request Sep 29, 2023
Draft
@andrewgait andrewgait requested a review from rowleya September 29, 2023 18:37
This was referenced Sep 29, 2023
Draft
Draft
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

@rowleya rowleya Awaiting requested review from rowleya

Assignees

No one assigned

Labels

enhancement

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

Ring buffer sigma and spikes-per-second on a per synapse-type basis

4 participants

@andrewgait @coveralls @rowleya @pabogdan