Use ndarray for matrix operations

[messense]

https://github.com/rust-ndarray/ndarray

[xd009642]

Just before I start looking into this is it still necessary (I notice the issue is >1 year old and code has obviously changed since then), and if so where in the code should I start looking 👀

[messense]

Not much has changed, I think it's the Context code.

crfs-rs/src/context.rs

Lines 26 to 100 in dcd02b5

	/// Context maintains internal data for an instance
	#[derive(Debug, Clone, Default)]
	pub struct Context {
	/// Flag specifying the functionality
	flag: Flag,
	/// The total number of distinct labels
	pub num_labels: u32,
	/// The number of items in the instance
	pub num_items: u32,
	/// The maximum number of labels
	cap_items: u32,
	/// Logarithm of the normalization factor for the instance.
	///
	/// This is equivalent to the total scores of all paths in the lattice.
	log_norm: f64,
	/// State scores
	///
	/// This is a `[T][L]` matrix whose element `[t][l]` presents total score
	/// of state features associating label #l at #t.
	pub state: Vec<f64>,
	/// Transition scores
	///
	/// This is a `[L][L]` matrix whose element `[i][j]` represents the total
	/// score of transition features associating labels #i and #j.
	pub trans: Vec<f64>,
	/// Alpha score matrix
	///
	/// This is a `[T][L]` matrix whose element `[t][l]` presents the total
	/// score of paths starting at BOS and arriving at (t, l).
	alpha_score: Vec<f64>,
	/// Beta score matrix
	///
	/// This is a `[T][L]` matrix whose element `[t][l]` presents the total
	/// score of paths starting at (t, l) and arriving at EOS.
	beta_score: Vec<f64>,
	/// Scale factor vector
	///
	/// This is a `[T]` vector whose element `[t]` presents the scaling
	/// coefficient for the alpha_score and beta_score.
	scale_factor: Vec<f64>,
	/// Row vector (work space)
	///
	/// This is a `[T]` vector used internally for a work space.
	row: Vec<f64>,
	/// Backward edges
	///
	/// This is a `[T][L]` matrix whose element `[t][j]` represents the label #i
	/// that yields the maximum score to arrive at (t, j).
	/// This member is available only with `CTXF_VITERBI` flag enabled.
	backward_edge: Vec<u32>,
	/// Exponents of state scores
	///
	/// This is a `[T][L]` matrix whose element `[t][l]` presents the exponent
	/// of the total score of state features associating label #l at #t.
	/// This member is available only with `CTXF_MARGINALS` flag.
	exp_state: Vec<f64>,
	/// Exponents of transition scores.
	///
	/// This is a `[L][L]` matrix whose element `[i][j]` represents the exponent
	/// of the total score of transition features associating labels #i and #j.
	/// This member is available only with `CTXF_MARGINALS` flag.
	exp_trans: Vec<f64>,
	/// Model expectations of states.
	///
	/// This is a `[T][L]` matrix whose element `[t][l]` presents the model
	/// expectation (marginal probability) of the state (t,l)
	/// This member is available only with CTXF_MARGINALS flag.
	mexp_state: Vec<f64>,
	/// Model expectations of transitions.
	///
	/// This is a `[L][L]` matrix whose element `[i][j]` presents the model
	/// expectation of the transition (i--j).
	/// This member is available only with `CTXF_MARGINALS` flag.
	mexp_trans: Vec<f64>,
	}