Skip to content

2016 08 08 Open NEST Developer Video Conference

Jump to bottom Edit New page
terhorstd edited this page Aug 21, 2016 · 6 revisions

previous | list | next

  1. Welcome

  2. Permanent archive of NEST Github material (discussion)

    • two kinds of materials:
      • repository (everyone has a clone, uncritical) and
      • discussions (issues, comments, etc.) only on github
    • github could take down the project without further notice, similar things happened e.g. with blogs on google
    • API exists, Dennis already has written a script to download issues and comments, machine will be installed for backups. SOLVED

  3. Removal of alias datatypes (issues #343 and #344; design decision)

    • data types like int_t, double_t

      • originally for easy substitution e.g. typedef float double_t; to half the memory consumption.
      • currently other memory structures much more limiting
      • at the moment there is not really a consistent use of double_t, so probably not really usable
      • suggestion: just do a global search&replace.

      DECISION: remove the typedefs and s&r the _t types

    • data types like port, weight, etc.

      •      + name carries information

      •      + eclipse shows comments on typedefs, very convenient

      •      - compiler will emit an unused variable warning, so var names are not an alternative

      •      - typedef to plain data-types does not provide type-checking, the impression of type-safety is misleading

      •      ? write classes with everything inline, constructors with explicit markers, reimplementing basic data types
        • mostly boilerplate, but unknown amount of debugging

      DECISION: alex will give this a try and we will get back to this in 2 weeks

  4. Removal of subnets in NEST 3.0 (issue #417; design decision)

    • subnets introduced long time ago

    • intended for different simulation resolutions

    • nowadays homogenious structures much better with bigger computers

    • subnets don't play any role, once the sim is running, basically only for net construction

    • gid collections much better mechanism, in future the create() command will return a gid collection

    • compact representation with iterator interface,

    • will make a lot of things a lot faster, also between python and c++ level

    • number of examples/test that need to be touched is low (20-30)

    • requires lots of users to change their scripts

    • plan is to do a 2.12 release with deprecation warnings, so users find infos how to convert their scripts

    • only point with real subnet usage: topology module, but HEP has solution

    • almost only advantages, but will break nest API

    • Q: gid collections should be accepted by functions taking gids? A: yes, this is the plan

    • C:

      • one branch for gidcollections
      • one branch for 2.12 development, deprecation warnings
      • we need people to change to NEST3 rather quickly, we don't want long transition a la python 2to3.

    DECISION: go this way.

  5. Review of NEST User Mailing List

    • approximting constatnt current : answered in last meeting
    • question about parallel computing : answered
    • seed of noise in the poisson ... : answered
    • Installation of NEST in SC St.diego : solved
    • Inibitory weight distribution :
      • log-normal distribution with negative sign in weights
      • suggestion to add a sign flag, will be handled by HEP directly
      • Q: is it possible to add arbitrary expressions that are still fast.
        • A: at present we only have the predefined set.

  6. Review of open Github Pull Request

    (wip: work-in-progress, Q A C: questions, answers comments)

    HIGHLIGHTED pull-requests:

    • #439: code looks ok, HEP will go through later this week

    • #429:

      • not understood why
      • Test on different machine shows different point of failure, if there are more threads than cores, but thread serialization must not make a difference.
      • travis tests break, Phillip (on holiday) maybe could help, but Jan will also look into this himself.

    • #448: version

      • 2.12-snapshot and 3.0-snapshot
      • using github hashes are not related to versions
      • using a suffix/odd number is more open for future verion decissisons
      • 2.11 would not require a suffix and follow general best practice

      DECISION: We use odd/even numbering and once a new version is near enough we can append a specific suffix (e.g. 3.0_rc1)

    further pull-requests:

    • #452: in contact with apyser

    • #451: reviewers assigned, wip

    • #447: HEP will look into this

    • #446: nest produces %.3f formatted output, for precise timing %.15f, if a certain precision is explicitly set, it should be respected. C: some of this will go away, as soon as nestIO is done

      wait till next meeting if there is some progress

    • #444: Eclipse, HEP dimitry

    • #436: probably ready for review soon.

    • #434: wip

    • #429: wip

    • #418: requires decision on allowed module names, HEP assigned

    • #387: ready to merge

    • #280: wip

    • #261: will converge within next few weeks, one principal issue: implement from paper or from private code?

      • e.g. basic membrane potential equation, original implementation: kernels started after time of spike. possible to reproduce paper by adding initial value

      DECISION: models should follow the papers.

    • #241: reviewed, author is on it. wait wip

    • #229: wip

    • #198: wip

  7. Review of open Github Issues

    not discussed due to lack of time.

    • decisions of this meeting as NESTDD (Design Decision) similar to Python PEPs?

      not discussed due to lack of time.

NEST Homepage: www.nest-simulator.org
NEST Initiative: www.nest-initiative.org

Status

  • License

  • DOI

Documentation

Open Video Meetings

User contributed Modules

Links

→ nest-initiative.org

→ nest-simulator.org

Clone this wiki locally