Struct CompressionParams

pub struct CompressionParams {
    pub pseudocounts: u64,
    pub topk: usize,
    pub structure_penalty: f64,
    pub topk_use_only_likelihood: bool,
    pub aic: f64,
    pub arity: u32,
}

Parameters for grammar induction.

Proposed grammars are scored as likelihood - aic * #primitives - structure_penalty * #nodes. Additionally, pseudocounts affects the likelihood calculation, and topk and arity affect what fragments can be proposed.

Fields

pseudocounts: u64

Pseudocounts are added to the observed counts associated with each primitive and invented expression.

topk: usize

Rather than using every expression in the frontier for proposing fragments, only use the topk best expressions in each frontier.

structure_penalty: f64

Structure penalty penalizes the total number of nodes in each Expression of the grammar’s primitives and invented expressions.

topk_use_only_likelihood: bool

Determines whether to use the maximum a-posteriori value for topk evaluation, or whether to use only the likelihood. Leave this to false unless you know what you are doing.

aic: f64

AIC is a penalty in the number of parameters, i.e. the number of primitives and invented expressions.

arity: u32

Arity is the largest applicative depth of an expression that may be manipulated to propose a fragment.

Trait Implementations

impl Default for CompressionParams

fn default() -> Self

The default params prevent completely discarding of primives by having non-zero pseudocounts.

CompressionParams {
    pseudocounts: 5,
    topk: 2,
    topk_use_only_likelihood: false,
    structure_penalty: 1f64,
    aic: 1f64,
    arity: 2,
}