Struct linfa_tsne::TSneParams

pub struct TSneParams<F, R>(_);

Implementations

impl<F: Float> TSneParams<F, SmallRng>

pub fn embedding_size(embedding_size: usize) -> TSneParams<F, SmallRng>

Create a t-SNE param set with given embedding size

Defaults to:

• approx_threshold: 0.5
• perplexity: 5.0
• max_iter: 2000
• rng: SmallRng with seed 42

impl<F: Float, R: Rng + Clone> TSneParams<F, R>

pub fn embedding_size_with_rng(
    embedding_size: usize,
    rng: R
) -> TSneParams<F, R>

Create a t-SNE param set with given embedding size and random number generator

Defaults to:

• approx_threshold: 0.5
• perplexity: 5.0
• max_iter: 2000

pub fn approx_threshold(self, threshold: F) -> Self

Set the approximation threshold of the Barnes Hut algorithm

The threshold decides whether a cluster centroid can be used as a summary for the whole area. This was proposed by Barnes and Hut and compares the ratio of cell radius and distance to a factor theta. This threshold lies in range (0, inf) where a value of 0 disables approximation and a positive value approximates the gradient with the cell center.

pub fn perplexity(self, perplexity: F) -> Self

Set the perplexity of the t-SNE algorithm

pub fn max_iter(self, max_iter: usize) -> Self

Set the maximal number of iterations

pub fn preliminary_iter(self, num_iter: usize) -> Self

Set the number of iterations after which the true P distribution is used

At the beginning of the training process it is useful to multiply the P distribution values by a certain factor (here 12x) to get the global view right. After this number of iterations the true P distribution value is used. If None the number is estimated.

Trait Implementations

impl<F: Float, R> ParamGuard for TSneParams<F, R>

fn check_ref(&self) -> Result<&Self::Checked, Self::Error>

Validates parameters

type Checked = TSneValidParams<F, R>

The checked hyperparameters

type Error = TSneError

Error type resulting from failed hyperparameter checking

fn check(self) -> Result<Self::Checked, Self::Error>

Checks the hyperparameters and returns the checked hyperparameters if successful

fn check_unwrap(self) -> Self::Checked

Calls check() and unwraps the result

impl<F: Float, R: Rng + Clone> Transformer<ArrayBase<OwnedRepr<F>, Dim<[usize; 2]>>, Result<ArrayBase<OwnedRepr<F>, Dim<[usize; 2]>>, TSneError>> for TSneParams<F, R>

fn transform(&self, x: Array2<F>) -> Result<Array2<F>>

impl<T, F: Float, R: Rng + Clone> Transformer<DatasetBase<ArrayBase<OwnedRepr<F>, Dim<[usize; 2]>>, T>, Result<DatasetBase<ArrayBase<OwnedRepr<F>, Dim<[usize; 2]>>, T>, TSneError>> for TSneParams<F, R>

fn transform(
    &self,
    ds: DatasetBase<Array2<F>, T>
) -> Result<DatasetBase<Array2<F>, T>>