Skip to content

Latest commit

 

History

History
177 lines (172 loc) · 8.02 KB

TODO.org

File metadata and controls

177 lines (172 loc) · 8.02 KB
Raw
  • AFTER 3.0, or in dedicated project +- Create examples for attaching external ASR/NLU/NLG and document it
    • vonda-nlp includes rasa NLU
    • vonda-minimqtt will include ASR using vosk (and maybe whisper?)
    • Create example for external HFC at run-time and StreamingClient connection
    • persistence in HFC: create directories, if missing?
  • RELEASE PLAN 4.0
    • Analyse von Java-Klassen z. automatischen Extraktion von Feld- und Methodensignaturen? Siehe jsignature, plus maven dependencies:list für den classpath
    • Sinnvolle Queries aus .rudi, deren Resultate gut verarbeitet werden können
    • could ‘>=’ be extended to java instanceof?
    • enum handling currently intransparant, it works if the enum is defined in the top-level Agent class, and there are no warnings, but why?

Bringt die baumartige Struktur überhaupt Vorteile? Alternativen?

Kann man die Generierung besser in eine eigene Stage verschieben?

die für sich genommen einfacher oder zumindest durchsichtiger ist?

Besseres plug-in für NLU/NLG/ASR/TTS über injection?

  • Die Funktionalität jetzt ist eigentlich ganz OK und funktioniert
  • rasa / Lucene / OpenNLP NLU als default-alternative
  • cplanner zur Luxus-NLG umbauen
  • Anschluss von vosk und anderen ASRs als modul, TTS?

Developer Tools?

  • Syntax-Highlighting f. VS Code? Anbindung an Rudibugger?

Queries aus .rudi abfeuern –> missing varargs!?

Collections probleme: Alles polymorph?

ich weiß nicht mehr, was damit gemeint ist

Dialogue Acts

  • hierarchy of Frames from the ontology with access functions
  • Nesting depth > 1 ?
  • TDL syntax statt CCG würg?

processor / abstract processor

  • behaviour sync
    • with optional delays ? (before/after)

Application examples

  • PAL system
  • Robo demo
  • HySociaTea SB demo
  • Herbea
  • Intuitiv

Softwarepraktikum

usability questionnaires / templates

  • which tools
  • how to debug
  • what works
  • what is awkward
  • report bugs

  • Timeouts mit “Argumenten”, die spezifische Instanzen beackern.
  • hierarchy of DAs
  • functionality: integrate jtfs for subsumption etc.
  • Final refactoring logging
    • Serialize DialogueAct to / from RDF
    • Magic Variable expansion in DialogueActs: slotnames and arguments with the same name as existing variables are replaced by the value of the variable, just as with {…} –> create an issue to REMOVE this //done
  • Implementation Patterns and Caveats
    • lastDAProcessed() before in propose: lastDA().theme :( //done
    • lastDAProcessed() after emitDA :( //done
    • in general: emitDA suspends the thread of commands!!! //done
    • if a question was asked: use waitingForResponse()
  • Fragen aus dem Durchforsten von VisitorType
    • ist dokumentiert, dass folgendes ein “clear” ist: //done child.name = null
    • done in 3.1.2, needs more detail! What turns into what.
    • Nicer strings for visualisation! More compact!
  • Release plan 3.0
    • check the option of not generating functions, and deferred generation in general
      • check the “final comments problem” and describe it
    • DEFNS is not defined by default!
      • This MUST be a variable in Agent.java! Default: ‘vonda’
      • This can then be changed using a config option, or by overwriting in toplevel generated file or wrapper class!
    • replace import by include
    • allow Java import statements at beginning of file only
      • bison grammar adapted
      • java imports have to be generated (currently ignored) + generation tested
      • Für .rudi files “include” statt “import” benutzen !!! JA FÜR 3.0
        • “import” wie in Java nur am Anfang des Files zulassen, dass dann auch nur Java imports macht?
    • VOnDA: VORSICHT, WENN MAN VARIABLEN BENUTZT, DIE HEIẞEN WIE PROPERTIES String name = “foo”; user.name = bla; // das greift auf user.foo zu!!! STEHT DAS IN DER DOKU?? WENN NICHT, SOFORT REINSCHREIBEN
      • Syntax ändern afu user.{name} ??
    • field access conflicts: solved
      • use “{exp}” syntax for property access, otherwise unexpected stuff happens.
      • Add it to AllYouCanDo.rudi and tests
    • get rid of reduce reduce conflicts
      • Übrige reduce-reduce Konflikte durch Syntax-Änderungen lösen, z.B. cast(TypeExpr, expr) statt (typeexpr) expr (oder isa statt cast)
      • simplest problem are the field/method defs, they could be later generated from a java parser/introspection
      • cast expressions –> `isa` infix operator: isa(type_expr, exp)
      • ARROW –> `lambda` infix operator: lambda(arg0, …) exp | body
    • Sollte man für die “if”s überhaupt Funktionen/Methoden generieren?
      • Wurde aufgehoben, so kann man das Problem mit der linearisierung von input und output code leichter lösen, see next
    • check the option of not generating functions, and deferred generation in general
      • check if break/return works as expected and write tests!!!
      • some things still move, since they need to be on top level, together with the process() function:
        • Field defs, to the front, but if they are combined with assignments, these will appear later
        • Method defs, they are currently deferred to the back
      • problem of “moving comments and injected java code” remains, but solution might be to pick comments BETWEEN two elements, is almost gone, but no guarantee due to moving code, and end-of-file situations, which can not be really fixed (code in or out of scope of process(){…}
      • check how good the comment handling really works
    • get rid of reduce reduce conflicts
      • remove ARROW token, unused
      • integrate with the rest
    • Improve the documentation
      • integrate new syntax changes
        • include and import
        • optional wrapper class
          • modify picture
        • “{exp}” syntax for field access
        • cast expressions
        • lambda expressions
        • type declarations for external fields/methods
    • Erleuchtung bzgl. Wrapperklasse: Agent vs. NaoAgent vs. PalAgent !!! JA FÜR 3.0 REALISIERT
      1. NaoAgent scheint eigentlich eine klasse wie TimelineUtils zu sein (oder könnte so sein)
      2. Sollten PalAgent und NaoAgent nicht die Plätze tauschen, also: NaoAgent extends PalAgent, und nicht umgekehrt.
      3. PalAgent wird nur einmal erzeugt, und alle Variablen und Methoden stehen also zur Verfügung, auch die nicht-statischen! Damit kann man aus Java-Klassen .rudi Methoden benutzen! Ausprobieren mit endQuiz (im Moment in NaoAgent)
    • Make wrapper class optional
      • Drawback if removed completely: you can not change functionality in Agent (framework), for example Behaviour handling. That is definitely a drawback! –> call it agent base class in config, and make it optional, if not given, extends from Agent(framework)
    • Modify special loading of type definitions: config param to specify name
      • except for Agent.rudi: seems to work fine!
    • Improve the documentation
      • check exisiting TODOs
      • describe the options with and without agentBase class
      • make clear that Java injection should be the ultimate exception
      • describe DEFNS and what it does
    • Finish converter
      • new fat jar target on branch
      • main conversion script
      • Type def file (sed!, or Java(no))
      • config file (sed, or Java, easy and safe)
        • wrapperClass -> agentBase + typeDef
        • class key for NLG (would speak in favour of java (embedding level 2)
    • StreamingClient problem persists, how to do that transparently
      • Currently, the “Agent” has to be registered externally: NO
      • RdfProxy registers itself on creation!
    • Introduce projectPrefix config key or init() parameter(?)
      • sets DEFNS and idPrefix
    • RELEASE PLAN FOR 3.0
      • Ask Jörg to re-iterate checks on M$
      • What can be done to check it’s really working (alpha first?)
        • vonda-chatcat: works
        • UnivFAQBot: works, errors fixed
        • VoluProf: works
    • Cheat Sheets Agent-Funktionen und Syntax
      • Überladene Operatoren: sind +=/-= und <= etc. die einzigen?
      • welche Vergleiche sind möglich
      • geht das auch für x is instance of Y: NEIN!