squiggle/packages/squiggle-lang/src/rescript/Distributions
2022-07-19 11:33:11 +10:00
..
PointSetDist Prior can't be a point 2022-07-12 09:45:41 -07:00
SampleSetDist Basic mapSampleN support 2022-07-19 11:33:11 +10:00
SymbolicDist Small beta error message change 2022-07-05 19:48:41 -04:00
DistributionOperation.res Formatted code 2022-07-13 10:11:59 -07:00
DistributionOperation.resi Moved env to GenericDist, and used it 2022-07-13 09:37:39 -07:00
DistributionTypes.res Prior can't be a point 2022-07-12 09:45:41 -07:00
GenericDist.res Formatted code 2022-07-13 10:11:59 -07:00
GenericDist.resi Formatted code 2022-07-13 10:11:59 -07:00
README.md Simplified tree in squiggle-lang/src/rescript/Distributions 2022-05-10 16:43:37 -04:00

Generic Distribution Library

This library provides one interface to generic distributions. These distributions can either be symbolic, point set (xy-coordinates of the shape), or sample set (arrays of random samples).

Different internal formats (symbolic, point set, sample set) allow for benefits and features. It's common for distributions to be converted into either point sets or sample sets to enable certain functions.

In addition to this interface, there's a second, generic function, for calling functions on this generic distribution type. This genericOperation standardizes the inputs and outputs for these various function calls. See it's run() function.

Performance is very important. Some operations can take a long time to run, and even then, be inaccurate. Because of this, we plan to have a lot of logging and stack tracing functionality eventually built in.

Diagram of Distribution Types

graph TD
    A[Generic Distribution] -->B{Point Set}
    A --> C{Sample Set}
    A --> D{Symbolic}
    B ---> continuous(Continuous)
    B ---> discrete(Discrete)
    B --> mixed(Mixed)
    continuous -.-> XYshape(XYshape)
    discrete -.-> XYshape(XYshape)
    mixed -.-> continuous
    mixed -.-> discrete
    D --> Normal(Normal)
    D --> Lognormal(Lognormal)
    D --> Triangular(Triangular)
    D --> Beta(Beta)
    D --> Uniform(Uniform)
    D --> Float(Float)
    D --> Exponential(Exponential)
    D --> Cauchy(Cauchy)

Diagram of Generic Distribution Types

Todo

  • Lots of cleanup
  • Simple test story
  • Provide decent stack traces for key calls in GenericOperation. This could be very useful for debugging.
  • Cleanup Sample Set library
  • Add memoization for calculations
  • Performance bechmarking reports
  • Remove most of DistPlus, much of which is not needed anymore
  • More functions for Sample Set, which is very minimal now
  • Allow these functions to be run on web workers
  • Refactor interpreter to use GenericDist. This might not be necessary, as the new reducer-inspired interpreter is integrated.

More todos