Add STDP synapse unit testing #1840

clinssen · 2020-11-12T16:46:29Z

This adds a unit test for the STDP synapse, by comparing the weight evolution due to a barrage of pre- and postsynaptic spikes between NEST and a Python reference implementation.

Test stimuli:

randomly generated (Poisson distribution) spike trains
hard-coded edge cases such as simultaneous pre and post spikes

Neuron models:

iaf_psc_exp
iaf_cond_exp
~~iaf_psc_exp_ps~~

Todo:

Unit test code cleanup
Add documentation

Addresses (fixes?) #1604.

clinssen · 2021-05-03T16:33:09Z

test_stdp_multiplicity has been fixed. The condition for test failure was not appropriate and rewritten. Performance is much better now. The following figure shows how, for smaller and smaller timestep, the final obtained weights of a "normal" versus a "precise" single STDP synapse converge.

Before (current master):

After (code in this PR):

heplesser · 2021-05-03T20:39:21Z

models/stdp_synapse.h

@@ -217,7 +217,7 @@ inline void
 stdp_synapse< targetidentifierT >::send( Event& e, thread t, const CommonSynapseProperties& )
 {
  // synapse STDP depressing/facilitation dynamics
-  const double t_spike = e.get_stamp().get_ms();
+  const double t_spike = e.get_stamp().get_ms() - e.get_offset();


@clinssen This change makes me wonder if we have the same error in other synapse models as well.

Yes, it does:

nest-simulator/models/clopath_synapse.h

Line 215 in 720f4c5

double t_spike = e.get_stamp().get_ms();

nest-simulator/models/ht_synapse.h

Line 189 in 720f4c5

const double t_spike = e.get_stamp().get_ms();

nest-simulator/models/jonke_synapse.h

Line 300 in 720f4c5

const double t_spike = e.get_stamp().get_ms();

nest-simulator/models/quantal_stp_synapse.h

Line 195 in 720f4c5

const double t_spike = e.get_stamp().get_ms();

nest-simulator/models/stdp_dopamine_synapse.h

Line 529 in 720f4c5

double t_spike = e.get_stamp().get_ms();

nest-simulator/models/stdp_nn_pre_centered_synapse.h

Line 235 in 720f4c5

double t_spike = e.get_stamp().get_ms();

nest-simulator/models/stdp_nn_restr_synapse.h

Line 229 in 720f4c5

double t_spike = e.get_stamp().get_ms();

nest-simulator/models/stdp_nn_symm_synapse.h

Line 231 in 720f4c5

double t_spike = e.get_stamp().get_ms();

nest-simulator/models/stdp_pl_synapse_hom.h

Line 244 in 720f4c5

const double t_spike = e.get_stamp().get_ms();

nest-simulator/models/stdp_synapse_hom.h

Line 281 in 720f4c5

const double t_spike = e.get_stamp().get_ms();

nest-simulator/models/stdp_triplet_synapse.h

Line 243 in 720f4c5

double t_spike = e.get_stamp().get_ms();

nest-simulator/models/tsodyks2_synapse.h

Line 209 in 720f4c5

const double t_spike = e.get_stamp().get_ms();

nest-simulator/models/tsodyks_synapse_hom.h

Line 256 in 720f4c5

const double t_spike = e.get_stamp().get_ms();

nest-simulator/models/urbanczik_synapse.h

Line 207 in 720f4c5

double t_spike = e.get_stamp().get_ms();

nest-simulator/models/vogels_sprekeler_synapse.h

Line 198 in 720f4c5

double t_spike = e.get_stamp().get_ms();

@clinssen So this means that until now, we have not taken spike time offsets into account in plasticity mechanisms, doesn't it? It does not have any effect on the actual timing of the spike transmitted, since the offset information is passed on to the receiving neuron in the Event object. Could you create a follow-up issue for this? We should involve @abigailm in the discussion.

heplesser

@clinssen This looks quite good, but some of the code should be simplified or better documented; tests need to be comprehensible ;).

pynest/nest/tests/test_stdp_multiplicity.py

pynest/nest/tests/test_stdp_synapse.py

heplesser · 2021-05-05T14:40:17Z

pynest/nest/tests/test_stdp_synapse.py

+            for self.nest_neuron_model in ["iaf_psc_exp",
+                                           "iaf_cond_exp",
+                                           "iaf_psc_exp_ps"]:
+                fname_snip = "_[nest_neuron_mdl=" + self.nest_neuron_model + "]"


What is this?

A snippet inserted into the filename when saving debug plots to disk.

heplesser · 2021-08-09T13:00:37Z

@clinssen Ping :)

clinssen · 2021-08-30T14:19:05Z

I removed all changes related to support for precise spike timing in synapses, and test coverage for precise neurons. This feature discussion will be continued in #2035.

heplesser

@clinssen Just a few more comments.

testsuite/pytests/test_stdp_multiplicity.py

testsuite/pytests/test_stdp_synapse.py

testsuite/pytests/test_stdp_nn_synapses.py

YounesBouhadjar

The PR looks good to me now, I just left a small comment

YounesBouhadjar · 2021-10-18T09:51:33Z

testsuite/pytests/test_stdp_synapse.py

+                idx_next_post_spike = np.where((post_spikes_delayed - t) > 0)[0][0]
+                t_next_post_spike = post_spikes_delayed[idx_next_post_spike]
+
+            if idx_next_post_spike >= 0 and t_next_post_spike < t_next_pre_spike:


t_next_post_spike or t_next_pre_spike might be referenced before assignment, if pre_spikes or post_spikes_delayed are empty. Maybe it's ok here, as we know that this never occurs in this specific test case.

YounesBouhadjar · 2021-10-18T10:04:31Z

testsuite/pytests/test_stdp_synapse.py

+        for self.dendritic_delay in [1., self.resolution]:
+            self.init_params()
+            for self.nest_neuron_model in ["iaf_psc_exp", "iaf_cond_exp"]:
+                fname_snip = "_[nest_neuron_mdl=" + self.nest_neuron_model + "]"


shouldn't you include the dendritic delay in the name here? otherwise, the corresponding figures of self.resolution will overwrite the figures of dendritic delay = 1.

jougs · 2021-10-21T08:51:40Z

@heplesser, can you please have another look? Thanks!

add STDP synapse unit testing

96f881a

clinssen marked this pull request as draft November 12, 2020 16:46

clinssen added I: No breaking change Previously written code will work as before, no one should note anything changing (aside the fix) S: Normal Handle this with default priority T: Enhancement New functionality, model or documentation labels Nov 23, 2020

clinssen added this to To do in Models via automation Nov 23, 2020

clinssen mentioned this pull request Jan 19, 2021

Unittest of dopamine synapse model #820

Open

clinssen moved this from To do to In progress in Models Mar 5, 2021

C.A.P. Linssen added 6 commits April 19, 2021 18:17

Merge remote-tracking branch 'upstream/master' into test-stdp-synapse

e3ac12a

allow weight_recorder to record precise spike times

75f3a4b

allow STDP synapse to use precise spike times

c8547d3

fix STDP synapse unit test

af21d05

pycodestyle formatting changes

a495dfe

Merge remote-tracking branch 'upstream/master' into test-stdp-synapse

791bff0

C.A.P. Linssen added 3 commits May 3, 2021 18:35

fix test conditions and add plots to test_stdp_multiplicity

a27b55c

fix Python code formatting

305e82c

clean up code in unit test (nest#1840)

ff08892

clinssen force-pushed the test-stdp-synapse branch 4 times, most recently from 3a5a5fa to adb0a5b Compare May 3, 2021 19:54

heplesser reviewed May 3, 2021

View reviewed changes

clinssen force-pushed the test-stdp-synapse branch 7 times, most recently from eaf2454 to 6c11fe7 Compare May 4, 2021 13:33

clinssen requested a review from YounesBouhadjar May 4, 2021 14:37

clinssen marked this pull request as ready for review May 4, 2021 14:37

heplesser requested changes May 5, 2021

View reviewed changes

Merge remote-tracking branch 'upstream/master' into test-stdp-synapse

4425620

C.A.P. Linssen added 2 commits August 30, 2021 13:52

clean up STDP multiplicity test

c878c9c

Merge remote-tracking branch 'upstream/master' into test-stdp-synapse

9c1cfa9

clinssen mentioned this pull request Aug 30, 2021

Add warning about synaptic plasticity mechanisms and precise spike timing #2137

Merged

C.A.P. Linssen added 2 commits August 30, 2021 15:10

move synaptic plasticity precise spike timing feature to nest#2035

fb9228d

refactor/clean up stdp synapse unit test

b4921bb

C.A.P. Linssen added 2 commits August 30, 2021 17:54

refactor/clean up stdp synapse unit testing

ae9f481

fix pycodestyle

7fb3e2d

heplesser requested changes Sep 6, 2021

View reviewed changes

C.A.P. Linssen added 4 commits October 4, 2021 12:26

Merge remote-tracking branch 'upstream/master' into test-stdp-synapse

ae60892

clean up unit tests following unittest/pytest migration

f4182fd

fix pycodestyle

14ad08c

clean up unit tests following unittest/pytest migration

c03797e

clinssen mentioned this pull request Oct 4, 2021

Document synaptic plasticity support for precise spike times #2176

Closed

clinssen requested a review from heplesser October 4, 2021 17:37

clinssen mentioned this pull request Oct 12, 2021

Concise and static Python imports #2135

Merged

YounesBouhadjar reviewed Oct 18, 2021

View reviewed changes

clean up based on review comments

7c7d949

clinssen requested a review from YounesBouhadjar October 20, 2021 10:42

YounesBouhadjar approved these changes Oct 20, 2021

View reviewed changes

heplesser approved these changes Oct 27, 2021

View reviewed changes

heplesser merged commit cfa80ce into nest:master Oct 27, 2021

heplesser moved this from In progress to Done in Models Nov 29, 2022

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Add STDP synapse unit testing #1840

Add STDP synapse unit testing #1840

clinssen commented Nov 12, 2020 •

edited

Loading

clinssen commented May 3, 2021

heplesser May 3, 2021

clinssen May 4, 2021

heplesser May 5, 2021

heplesser left a comment

heplesser May 5, 2021

clinssen Aug 30, 2021

heplesser commented Aug 9, 2021

clinssen commented Aug 30, 2021

heplesser left a comment

YounesBouhadjar left a comment

YounesBouhadjar Oct 18, 2021 •

edited

Loading

YounesBouhadjar Oct 18, 2021

jougs commented Oct 21, 2021

Add STDP synapse unit testing #1840

Add STDP synapse unit testing #1840

Conversation

clinssen commented Nov 12, 2020 • edited Loading

clinssen commented May 3, 2021

heplesser May 3, 2021

Choose a reason for hiding this comment

clinssen May 4, 2021

Choose a reason for hiding this comment

heplesser May 5, 2021

Choose a reason for hiding this comment

heplesser left a comment

Choose a reason for hiding this comment

heplesser May 5, 2021

Choose a reason for hiding this comment

clinssen Aug 30, 2021

Choose a reason for hiding this comment

heplesser commented Aug 9, 2021

clinssen commented Aug 30, 2021

heplesser left a comment

Choose a reason for hiding this comment

YounesBouhadjar left a comment

Choose a reason for hiding this comment

YounesBouhadjar Oct 18, 2021 • edited Loading

Choose a reason for hiding this comment

YounesBouhadjar Oct 18, 2021

Choose a reason for hiding this comment

jougs commented Oct 21, 2021

clinssen commented Nov 12, 2020 •

edited

Loading

YounesBouhadjar Oct 18, 2021 •

edited

Loading