Skip to content

Latest commit

 

History

History
83 lines (67 loc) · 4.27 KB

L21 Configuration.md

File metadata and controls

83 lines (67 loc) · 4.27 KB

Definitions

Design

  • Design:
    • input can be a need/goal/function
    • output: specification/structure of some artifact that satisfies the input
    • can be open-ended or well-defined
  • Design vs problem-solving
    • Problem-solving: typically, the problem remains fixed (even the solution evolves)
    • Design: the problem and the solution co-evolve

Configuration

  • Configuration:

    • an arrangement of all components
    • a problem-solving activity that assigns values to variables to satisfy constraints
    • Which variables should we set first? Different heuristics for the ordering of variables:
      • set the most restricted variables first
      • set the variables that restrict others the most first
      • set the most important variables first

  • Configuration vs design

    • Configuration: a type of design that is routine. All components are known. You simply arrange the components
    • Design: components not known

  • Example: arranging furniture in a living room

    • Constraints: e.g. TV must be placed in front of couch; distance between TV and couch, etc.
    • Most important: location of TV, then location of couch
      • Therefore, set the TV first, then the couch, then other items

The configuration process

  • Start with specifications of all the constraints
  • Output: an arrangement model
  • As shown below, one may begin with some abstract/partial solutions: a design plan
    • assign values to variables in the plan
    • refine and expand the plan
      • the plan specifies more variables at lower-level
      • assign values to variables in the plan
  • repeat the above steps until we have a complete arrangement model (solution)
  • Assess the arrangement model. If needed, repeat the process
  • e.g. the runtime of the configuration comes out to be slower than the specifications. Two solutions:
    • change the specifications of the problem (i.e. solution and problem co-evolve)
    • repeat the process until the constraints (e.g. runtime) are satisfied

  • Abstraction hierarchy: plans arranged at different levels of abstraction
  • Component hierarchy: In configuration, all components are known, but a component can belong to a specific kind or a general kind. Sometimes you can select the kind you want in configuration.

Example: representing a chair

  • Figure above: a chair represented with frames (prototype of a chair)
  • Configuration design: figure out the arrangement of all the components of a chair by filling out the slots of the frames
  • Values of slots (variables) must satisfy global constraints (see red box) and the range of values that a slot can take
  • Different designers use different design plans. In the following example:
    • the abstract plan has a cost range for each component (<$5) (may change later)
    • the count of chair legs is assigned first before other variables
    • Expand the plan after defining the count of chair legs (e.g. deal with more detailed constraints such as the size and material of chair legs; this is an example of constraint propagation(Lesson 20))

Comparisons with classification/ case-based reasoning/ planning

Connection to classification

  • Classification and configuration are both hierarchical
    • Classification: "establish and refine" (Lesson 11)
    • Configuration: initiate a plan, than refine and expand the plan
  • Configuration leverages classification's notion of prototype concepts
    • e.g. in the chair example above, configuration starts with a plan representing the prototype of a chair (which provides all the components of a chair)
    • (In other words, classification helps us to make sense of the world, configuration helps us to create a world)

Contrast with case-based reasoning

  • Configuration: starting with a prototype concept and assigning values to variables
  • Case-based reasoning: starting from a specific chair and tweaking it as needed
  • (We will learn how an AI agent decides which of the above methods to use)

Connection to planning

  • The result of a planning task can lead to a prototype that can subsequently be configured for similar problems with differing constraints.

Cognitive connection

  • Daily routines, e.g. cooking
  • Separation of tasks and methods: configuration is a task that can be addressed by many methods (e.g. planning, case-based reasoning, etc.)