imprecisegen

Work in progress!

Implements imprecise probabilities in Gen.jl. Currently only supports enumerative inference.

Files

• knightian.jl defines a new probability distribution allowing Knightian choices from distributions. An error is thrown if one attempts to sample from it.
• imprecise-enumerative.jl implements a modified form of enumerative inference in which samples and Knightian choices are separated.
• lib.jl defines functions for convenient manipulation of imprecise probabilities.
• visualize.jl allows for visualization of the resulting distributions as convex regions.
• examples
  • cip.jl implements the examples from Compositional Imprecise Probability by Jack Liell-Cock and Sam Staton.
  • worldmodel.jl has a simple Bayesian credal set update example right now, but will have a more complicated example of an agent's model
  • predictions.jl will be for showing how we can specify problems like P(at least two of A, B, C occur) given known probabilities P(A), P(B), P(C)
  • twourn.jl will be for ambiguity aversion problems from economics with maxmin utility theory

Dependencies

• Gen.jl, most recent committed version on GitHub for using enumerative inference (will switch to an official version once it's published)
• IterTools
• Plots.jl and LazySets.jl for visualization

TODO

• automatically enumerate choices in generative function rather than requiring them to be manually specified - this will simplify the code of further examples considerably
• finish the examples above