Struct ppca::PPCAMix

pub struct PPCAMix(_);

A mixture of PPCA models. Each PPCA model is associated with a prior probability expressed in log-scale. This models allows for modelling of data clustering and non-linear learning of data. However, it will use significantly more memory and is not guaranteed to converge to a global maximum.

Notes

• The list of log-weights does not need to be normalized. Normalization is carried out internally.
• Each PPCA model in the mixture might have its own state size. However, all PPCA models must have the same output space. Additionally, the set of PPCA models must be non-empty.

Implementations

impl PPCAMix

pub fn new(models: Vec<PPCAModel>, log_weights: DVector<f64>) -> PPCAMix

pub fn init(n_models: usize, state_size: usize, dataset: &Dataset) -> PPCAMix

Creates a new random untrained model from a given number of PPCA modes, a latent state size, a dataset and a smoothing factor. The smoothing factor helps with overfit of rarely occuring dimensions. If you don’t care about that, set it to 0.0.

pub fn output_size(&self) -> usize

The number of features for this model.

pub fn state_sizes(&self) -> Vec<usize>

The number of hidden values for each PPCA model.

pub fn n_parameters(&self) -> usize

The total number of parameters involved in training (used for information criteria).

pub fn models(&self) -> &[PPCAModel]

The list of constituent PPCA models.

pub fn log_weights(&self) -> &DVector<f64>

The log-strength (or log-a priori probablilty) for each PPCA model.

pub fn weights(&self) -> DVector<f64>

The strength (or a priori probablilty) for each PPCA model.

pub fn sample(&self, dataset_size: usize, mask_probability: f64) -> Dataset

Sample a full dataset from the PPCA model and masks each entry according to a Bernoulli (coin-toss) distribution of proability mask_prob of erasing the generated value.

pub fn llk_one(&self, sample: &MaskedSample) -> f64

Computes the log-likelihood for a single sample.

pub fn llks(&self, dataset: &Dataset) -> DVector<f64>

Computes the log-likelihood for each sample in a dataset.

pub fn llk(&self, dataset: &Dataset) -> f64

Computes the total log-likelihood for a given dataset.

pub fn infer_cluster(&self, dataset: &Dataset) -> DMatrix<f64>

Returns the posterior distribution (i.e., with Baye’s rule applied) for each sample in the given dataset. Each row of the matrix corresponds to a categorical distribution on the probability of a sample belonging to a particular PPCA model.

pub fn infer_one(&self, sample: &MaskedSample) -> InferredMaskedMix

Infers the probability distribution of a single sample.

pub fn infer(&self, dataset: &Dataset) -> Vec<InferredMaskedMix>

Infers the probability distribution of a given dataset.

pub fn smooth_one(&self, sample: &MaskedSample) -> MaskedSample

Filters a single samples, removing noise from it and inferring the missing dimensions.

pub fn smooth(&self, dataset: &Dataset) -> Dataset

Filters a dataset of samples, removing noise from the extant samples and inferring the missing samples.

pub fn extrapolate_one(&self, sample: &MaskedSample) -> MaskedSample

Extrapolates the missing values from a sample with the most probable values.

pub fn extrapolate(&self, dataset: &Dataset) -> Dataset

Extrapolates the missing values from a dataset with the most probable values.

pub fn iterate(&self, dataset: &Dataset) -> PPCAMix

Makes one iteration of the EM algorithm for the PPCA mixture model over an observed dataset, returning a improved model. The log-likelihood will always increase for the returned model.

pub fn iterate_with_prior(&self, dataset: &Dataset, prior: &Prior) -> PPCAMix

Makes one iteration of the EM algorithm for the PPCA mixture over an observe dataset, using a supplied PPCA prior (same for all constituent PPCA models) and returning the improved model. This method will not necessarily increase the log-likelihood of the returned model, but it will return an improved maximum a posteriori (MAP) estimate of the PPCA model according to the supplied prior.

pub fn to_canonical(&self) -> PPCAMix

Maps PPCAModel::to_canonical for each constituent model.

Trait Implementations

impl Clone for PPCAMix

fn clone(&self) -> PPCAMix

Returns a copy of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for PPCAMix

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl<'de> Deserialize<'de> for PPCAMix

fn deserialize<__D>(__deserializer: __D) -> Result<Self, __D::Error>where
    __D: Deserializer<'de>,

Deserialize this value from the given Serde deserializer.

impl Serialize for PPCAMix

fn serialize<__S>(&self, __serializer: __S) -> Result<__S::Ok, __S::Error>where
    __S: Serializer,

Serialize this value into the given Serde serializer.