PragmaticGP_pile2.feature

Feature: random Pile script generator
  In order to get sample arbitrary Pile solutions to a target problem
  As a symbolic regression modeler
  I want to be able to compose random Pile scripts, sampled from a set of tokens

  Scenario: sample available tokens uniformly
    When I generate 10000 random Pile scripts, each with 20 tokens
    And the available tokens are ["a", "b", "c"]
    Then the tokens should appear (on average) with equal frequency
    And all possible pairs should appear (on average) with equal frequency

  Scenario: random constants use "k" as a placeholder
    When I generate 10000 random Pile scripts, each with 20 tokens
    And the available tokens are ["k"]
    Then every script should contain 20 floating-point numbers
    And every generated number should have at most 3 digits after the decimal
    And every generated number should be within [-20.000,20.000]

  Scenario: should be able to generate scripts of a given size
    When I generate 10 random Pile scripts, each with 100 tokens
    Given the available tokens are ["foo", "bar", "baz", "qux", "hah"]
    Then every script should contain 100 tokens
    And every script should contain 399 characters
    And every script should contain 99 spaces