Discussion

  • 2 d  discrete space doesn’t seem to be very comfortable with Newton’s F = Gm1m2/r2. So, I am looking at
    • scaling space so that I can use real number for G, distances, etc. Not sure about this.
    • Seeing if there is a different ruleset for a discrete  2-d context
  • Meanwhile, the modeling environment seems to be working ok. I am adding needed  features as we go along, e.g., a graphics viewing capability, add / deleting agents, debugging frames, test programs, …
  • Teething problems with infrastucture – vpython for 3d visualizations, versioning conflicts, …
Advertisements

Code Snapshot pb4

Code with multiple dimensions working (as far as I can tell) Continue reading

Ruleset

  • Ruleset ideas – specification language:
    • agents have attributes, location in n-space
    • there are rules for how these attributes change, change the agents direction, how they interact
    • creation of another dimension – a group becomes an agent in the next higher dimension following the rules of that dimension
    • initial conditions specs:
      • population – number of agents
      • number of starting dimensions
      • size of each dimension
      • conditions that add a dimension either for just an agent, for a group, or for the entire space
      • data that needs to be collected, how frequently (at each step, after n steps, …)
      • data destination – file, graphics, ….

Discussion

  • simulation of 2 body, 3 body, n body problem in 2 d, 3 d, n d  as a starting point to flush out the modeling environment, develop some features? How does this work in discrete space as opposed to continuous space? Does n space change the parametrization?
  • Ruleset ideas design
  • What is space any way ? Coordinate space, configuration space, attribute space, memetic space ? Is there an underlying  notion of space of which these are all examples / tranformations?

Post By Email Shortcodes

Shortcodes to use in email for posting by email

Special shortcodes can be embedded in your email to configure various aspects of the published post: Continue reading

Do List

Add features:

  • Shadow capability – projection of higher dimension in to lower dimension, tracking parent self, modifying parent behavior, …?
  • Group information gathering – number of groups, sizes of groups, changes in groups(?), frequency of group data collections (every step, n steps?)
  • Include moore (diagonals) neighborhood capability
  • include center capability (off/on) in neighborhood detections
  • Multiple or single occupancy per cell selection (in dimensions.py / place_agent)
  • Profile infrastructure to track performance
    • automate profiling
    • develop meta model profiling – standard, metrics, number of steps, dimensions, etc.
    • easy access to profiling history
  • Use generators for performance improvement and benchmark profile before and after to see what difference they make
  • Change blog to accept email input for posting
  • Version control – set up for model, think up a release / version system

Modeling Discussion List

  • Need mathematica – upgrade to 11.3 ?
  • Share / open source modeling features ?
  • Blog set up to track work, profiling, history
  • Shadows – projection into lower dimensions, all permutations?
  • Visualization – charts, projections, what data to collect
  • Single / multiple occupancy per cell
  • Ruleset interface
  • Gravity field ? Calculate effect on each agent of every other agent G = Gm1m2/r**2 rather than agent affect directly on another agent, i.e., agent affects context and then context affects agent

Python libraries

  • Calling any external program, in a subprocess, and returning the results to a Python program — but with the same syntax as if the program in question were a native Python function

Cuda Refs

Cuda compute capabilities

cuda vs openCL

Agent Based Modeling Tools

Flame gpu  uses cuda to do agent based modeling