2016 08 22 Open NEST Developer Video Conference

previous | list | next

1. Welcome

2. NEST User Workshop, Karlsruhe, Germany, 3/4 November 2016

   • VC on Thursday to discuss agenda
     • few talks/presentations (API changes, NEST I/O, multi-compartment NEST advertisement, ...)
     • split in groups of hands on (mostly robotics, NEST hacking, ...)
     • Annual meeting of the NEST Initiative

3. Some numerical experiences

   1. PR 466

      • regressiontest fails: two neurons constant current, record spikes of second neuron membrane potential

        • completely different results on the IBM machines

        • everywhere spikes 50, 100, 150, ...

        • IBM 50.1, 100.2, ....

        • traced down to calculation of current for spiking frequency here

          /I0 Theta C mul Tau div 1.0 1.0 f div TauR sub Tau div neg exp sub div def

      • found one bit difference with IBM library, cross-checked with mathematica, IBM solution is correct
      • rewritten the test, to be numerically more stable
      • need to be aware that one bit noise can have significant impact on simulation results
      • conclusion: if you need precisely timed spikes, don't use constant-current inputs

   2. PR 229 how to handle exploding exponential in a numerically correct way?

      • few proper papers/references
      • LSODAR solver availble with python wrapers, takes threshold conditions into account,
      • a reference solution was set up, spikes now correct, adaptive step size much smaller around spikes (good), tests with different bounding of the argument of the exponential set up (not the correct way to go, but gives better results for now)
      • conclusion: tools to investigate this behaviour are at hand, but LSODAR is not re-entrant so not usable from NEST.

4. Travis problems (see Travis bug report)

   • travis tests with MPI fail due to obscure graphics driver problem
   • @jougs is in contact with travis people
   • needs to be fixed rather soon, because PRs can not be green before

5. Python-based help generation (Design Decision, see #464)

   • @steffengraber has reimplemented parser in python,
   • concern: machines without python cannot create userdoc (e.g. K) -> non-problem
   • doctest-like function in SLI, only for round(), fails if help-file not generated. no real problem.

   DECISION: move doc generation to python

6. Review of NEST User Mailing List

   • all emails answered, all points addressed

7. Review of open Github Pull Request (wip = work in progress)

   • PR 467: @apeyser works with RWTH -> @skunkel
   • PR 466: -> @DimitriPlotnikov
   • PR 463: postpone till @hannahbos is back
   • PR 464: was agreed on, review ongoing
   • PR 462: only one word change -> assign to @heplesser
   • PR 449: ready to merge (see what github does regarding PR 462 with same change)
   • PR 454: needs travis back up, essentially approved
   • PR 451: wait for @mschmidt87 and @jhnnsnk
   • PR 448: version 2.11 -> @DimitriPlotnikov
   • PR 446: needs travis back up
   • PR 444: eclipse, wip
   • PR 439: needs travis back up additional review -> @DimitriPlotnikov
   • PR 436: -> @apeyser
   • PR 434: needs travis back up
   • PR 429: needs travis back up (needs MPI testing) -> @skunkel Interesting: found race-conditions with intel parallel debuging tools, could be false-positives, tools could be useful.
   • PR 418: load modules at startup, wip
   • PR 280: ongoing communication, wip
   • PR 261: @heplesser looks into it
   • PR 241: wip
   • PR 229: talked about (see above)
   • PR 198: wip, only formatting issues, put #action-required

8. Review of open Github Issues

   • out of time / not discussed