Add iaf_psc_exp_ps_lossless model by Krishnan et al #901

heplesser · 2018-03-07T16:12:09Z

This PR is a follow-up to the retired #648. Two issues remain at present:

The test for the model fails, because the reference data are taken from the non-precise iaf_psc_exp model. We need reference data generated outside of NEST for this model.
The test for the model is very simple, just response to DC input; it does not test any of the advanced features, in particular the losslessness.
There is some problem with the copyright header, probably a technical problem with the test.

…hms on precise version of exp-neuron in

This reverts commit 3ae028f.

# Conflicts: # precise/iaf_psc_alpha_canon.cpp # testsuite/unittests/test_dcgen_versus_I_e.sli

heplesser · 2018-03-09T10:45:01Z

I encountered a problem with the issue-368.sli regression test related to #368. When the model receives an excitatory spike (which does not evoke a spike on its own), and an inhibitory spike at the same time, then neuron will emit a spike in the past. This happens if we do not accumulate incoming spikes, which we should never do, as discussed in #368.

This needs to be resolved before merging!

ChristianKeup · 2018-03-09T15:14:55Z

Hi, so I was assigned by Moritz to help completing this pull request.
Since this is my first adventure into the nest code and I don't know SLI, pls be patient with me!
I would like to know what exactly you want me to do:

fix the unittest
expand the unittest to incorporate more features -> which ones?
try to fix the problem related to Incorrect summation of synaptic currents for iaf_psc_exp_ps #368 just mentioned ?
more?

Also, do you want me to work directly with (i.e. push to) the heplesser:krishnanj-time-reversal-precise branch, or should I use my own branch that will then in the end be merged into nest:master? (sorry if this is clear)

heplesser · 2018-03-10T12:00:43Z

@ChristianKeup Welcome aboard! You summarized the remaining tasks quite well. Some comments:

Currently, three tests are failing. We will work on the problem with ticket-659-copyright, that seems to be some strange technicality.
The issue-77.sli test is failing due to the problem related to Incorrect summation of synaptic currents for iaf_psc_exp_ps #368; I will also try looking into it, but it would be good if you could take a look as well. As I wrote before, the problem is caused by spikes arriving at the same time with opposite signs.
test_iaf_psc_exp_ps_lossless.sli fails because the reference values stored in the file are not correct. Only those values need to be updated and replaced by correct values that have been obtained differently. Since in that test the model is only driven by constant input, it should be possible to obtain the correct solution analytically.
What is missing are tests that test whether the "new stuff" in the model works, i.e., the guaranteed detection of spikes also in corner cases. Does the paper contain examples of such cases we could use for the test? If you prefer, you could implement the extended tests in Python.

Concerning the practical procedure: This PR is from heplesser:krishnanj-time-reversal-precise, i.e., branch krishnanj-time-reversal-precise in my Github fork of NEST (https://github.com/heplesser/nest-simulator). You should check out that branch, commit your changes to your local copy of the branch, the push it to your NEST fork on Github and create a PR against my branch (not master). I can then integrate your changes and they will end up here in the PR against master.

stinebuu · 2018-03-12T10:06:26Z

@heplesser concerning ticket-659-copyright, there seems to be a whitespace after * test_iaf_psc_exp_ps_lossless.sli in the test file. Removing the whitespace makes the copyright error disappear.

heplesser · 2018-03-14T14:43:26Z

@stinebuu Thanks for the detective work!

ChristianKeup · 2018-03-19T16:33:15Z

I have found a problem where the lossless neuron is behaving for small inputs close to threshold like a normal unprecise ias_psc_exp neuron. I'm currently still looking into it.

…ive simultaneously.

heplesser · 2018-03-20T14:39:15Z

@ChristianKeup I have just pushed a change that solves the problem with issue-77.sil by explicitly skipping state propagation and spike check if two spikes arrive at the same time, so that ministep==0. I don't know if that will help the issue you are working on. The problem with the copyright check is also fixed.

ChristianKeup · 2018-03-20T16:27:54Z

precise/iaf_psc_exp_ps_lossless.cpp

+          if ( not numerics::is_nan( spike_time_max ) )
+          {
+            emit_spike_( origin, lag, V_.h_ms_ - last_offset, spike_time_max );
+          }
        }



@heplesser thanks!
If I see it correctly, the issue of a ministep == 0 is handled slightly differently in iaf_psc_exp_ps.cpp, where this is done by checking that dt != 0 in propagate_( dt ) and emit_spike( ... , dt ).
I think it does the same, but just wanted to mention this in case you didn't do it differently on purpose

@ChristianKeup Ah, I overlooked that. I will update my code.

…pike_()

…ossless

ChristianKeup · 2018-03-30T16:15:38Z

The problem I mentioned turned out to be due to external currents, as from a /dc_generator, not being taken into account in the spike detection method is_spike( dt ). It only took into account the external current given by the model parameter P_.I_e_. (Is it on purpose that the two are treated seperately?)

In the process I checked the equations implemented in is_spike( dt ). They are in fact correct and working. However, it turned out that eq. (49) in Krishnan et al. 2018 is not correct. Also, in the spike detection algorithms 1 and 2 in the paper, the functions g and f are interchanged. (I have mailed with Luca about this)
I changed to code to be less misleading and added a short comment in case someone else should ever try to understand the implementation.

Now the model is working nicely. I fixed the current unittest and also added the exact same unittest for the lossy /iaf_psc_exp_ps since none yet existed. In this simple test, both models behave exactly the same, as they should. I will write an extension testing for the region of previously missed spikes next week.
@heplesser I've created a PR with the changes against your branch, hopefully in the correct way.

Changes to implementation and test of iaf_psc_exp_ps_lossless for Nest PR nest#901

…implemented.

heplesser · 2018-04-07T13:16:50Z

@Silmathoron @stinebuu I believe this PR is now in order. Could you have another look? For background on some of the changes and design decisions, please see also the discussion on heplesser#22

stinebuu

I found a tiny typo, otherwise it looks good.

stinebuu · 2018-04-09T11:23:19Z

precise/iaf_psc_exp_ps_lossless.cpp

+    }
+    else
+    {
+      // We only get here if there is at least on event,


I am guessing it is supposed to be 'one' and not 'on'?

@stinebuu Fixed.

Fixed typo in comment.

Silmathoron

All good for me

krishnanj and others added 30 commits January 23, 2017 13:31

models implementing time-reversal state space spike detection algroit…

1ccdbc7

…hms on precise version of exp-neuron in

associated variables for time reversal model included in nestkernel

891f591

added description for the time reversal model

d865f90

modify function description

564a5b8

removed temp vars

d1f19dc

more comments

17b73e1

addressing hep's comments

49e6ff4

addressing hep's comments

4194ad7

addressing hep's comments

22ca54d

addressing hep's comments

22b7815

addressing hep's comments

969272e

addressing hep's comments

2423fa0

addressing hep's comments

da1ae0d

addressing hep's comments

171a858

addressing hep's comments

0bda5c4

addressing hep's comments

a759639

addressing hep's comments

4c53f42

addressing hep's comments

5744488

addressing hep's comments

0c98af5

code formatting issues

53c2f7c

code formatting

5080ccd

handle get_next_spike

e044cfe

handle next spike

d1b53fe

resolve conflicts

3ae028f

addressing hep's comments; add unit test

9ad0712

Revert "resolve conflicts"

5955541

This reverts commit 3ae028f.

Restore slice_ring_buffer.h.

40ec4aa

Fixes for iaf_psc_exp_ps_lossless model.

48955e3

Merge branch 'master' into krishnanj-time-reversal-precise

db0e3d7

# Conflicts: # precise/iaf_psc_alpha_canon.cpp # testsuite/unittests/test_dcgen_versus_I_e.sli

Fixed inconsistent E_L initialization.

7beb90a

Removed trailing whitespace.

effae6c

Skip membrane potential propagation and spike check if two spikes arr…

c9aeeab

…ive simultaneously.

ChristianKeup reviewed Mar 20, 2018

View reviewed changes

heplesser and others added 4 commits March 21, 2018 13:41

Also iaf_psc_exp_ps now skips propagation by branch in update().

5f6470f

Handle external dc current correctly in iaf_psc_exp_ps_lossless::is_s…

9480e8e

…pike_()

resolves confusion of g and f in spike detection of /iaf_psc_exp_ps_l…

f8a7a62

…ossless

fixed test_iaf_psc_exp_ps_lossless and added test_iaf_psc_exp_ps

0b1ef91

ChristianKeup mentioned this pull request Mar 30, 2018

Changes to implementation and test of iaf_psc_exp_ps_lossless for Nest PR #901 heplesser/nest-simulator#22

Merged

ChristianKeup added 3 commits April 5, 2018 14:33

adressed formatting issues, comment

e0cda00

extended test_iaf_psc_exp_ps_lossless.sli to check spike detection alg.

d8cbb64

again adressing clang-format issues

d5e66e5

ChristianKeup mentioned this pull request Apr 6, 2018

Allow different time constants for exc. and inh. synapses in /iaf_psc_exp_ps_lossless #921

Closed

ChristianKeup and others added 4 commits April 6, 2018 17:29

enforced tau_syn_ex == tau_syn_in for /iaf_psc_exp_ps_lossless

ec7586e

Merge pull request #22 from ChristianKeup/lossless

49fd9c9

Changes to implementation and test of iaf_psc_exp_ps_lossless for Nest PR nest#901

Improved error message and fixed formatting.

52a541b

Test ticket-686-... ignores iaf_ps_exp_ps_lossless until nest#921 is …

703d138

…implemented.

stinebuu approved these changes Apr 9, 2018

View reviewed changes

Update iaf_psc_exp_ps_lossless.cpp

4a39c27

Fixed typo in comment.

Silmathoron approved these changes Apr 13, 2018

View reviewed changes

heplesser merged commit e24b13f into nest:master Apr 13, 2018

heplesser deleted the krishnanj-time-reversal-precise branch April 19, 2018 12:10

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Add iaf_psc_exp_ps_lossless model by Krishnan et al #901

Add iaf_psc_exp_ps_lossless model by Krishnan et al #901

heplesser commented Mar 7, 2018

heplesser commented Mar 9, 2018

ChristianKeup commented Mar 9, 2018

heplesser commented Mar 10, 2018

stinebuu commented Mar 12, 2018

heplesser commented Mar 14, 2018

ChristianKeup commented Mar 19, 2018

heplesser commented Mar 20, 2018

ChristianKeup Mar 20, 2018

heplesser Mar 21, 2018

ChristianKeup commented Mar 30, 2018 •

edited

Loading

heplesser commented Apr 7, 2018

stinebuu left a comment

stinebuu Apr 9, 2018

heplesser Apr 9, 2018

Silmathoron left a comment

Add iaf_psc_exp_ps_lossless model by Krishnan et al #901

Add iaf_psc_exp_ps_lossless model by Krishnan et al #901

Conversation

heplesser commented Mar 7, 2018

heplesser commented Mar 9, 2018

ChristianKeup commented Mar 9, 2018

heplesser commented Mar 10, 2018

stinebuu commented Mar 12, 2018

heplesser commented Mar 14, 2018

ChristianKeup commented Mar 19, 2018

heplesser commented Mar 20, 2018

ChristianKeup Mar 20, 2018

Choose a reason for hiding this comment

heplesser Mar 21, 2018

Choose a reason for hiding this comment

ChristianKeup commented Mar 30, 2018 • edited Loading

heplesser commented Apr 7, 2018

stinebuu left a comment

Choose a reason for hiding this comment

stinebuu Apr 9, 2018

Choose a reason for hiding this comment

heplesser Apr 9, 2018

Choose a reason for hiding this comment

Silmathoron left a comment

Choose a reason for hiding this comment

ChristianKeup commented Mar 30, 2018 •

edited

Loading