Struct markovr::MarkovChain

pub struct MarkovChain<T: Element> { /* fields omitted */ }

Variable-order Markov chain.

Implementations

impl<T: Element> MarkovChain<T>

pub fn new(order: usize) -> Self

Creates a new MarkovChain.

'order' is the order of the Markov Chain.

A value of 1 is your typical Markov Chain, that only looks back one place.

A value of 0 is just a weighted die, since there is no memory.

You could think of order as the shape of the input tensor, where the input tensor is the sliding view / window.

pub fn train(&mut self, view: &[T], result: T, weight_delta: i32)

Feeds training data into the model.

'view' is the sliding window of elements to load into the MarkovChain chain. the number of elements in view should be self.order + 1 (excess will be ignored). the last element in view is the element to increase the weight of.

'weight_delta' should be the number of times we're loading this view into the model (typically 1 at a time).

pub fn generate_deterministic(&self, view: &[T], rand_val: u64) -> Option<T>

Generates the next value, given the previous item(s).

view is the sliding window of the latest elements. only the last self.order elements are looked at.

rand_val allows for a deterministic result, if supplied.

pub fn generate(&self, view: &[T]) -> Option<T>

Generates the next value, given the previous item(s).

view is the sliding window of the latest elements. only the last self.order elements are looked at.

pub fn probability(&self, view: &[T], result: T) -> f32

Returns the probability of getting 'result', given 'view'.