GitHub

Mixture Models

A mixture model is a probability distribution which, given a finite $k > 0$, samples from $k$ different distributions $\{f_i(x) | i \in \{1,...,k\}\}$ randomly, where the probability of sampling from $f_i(x)$ is $\pi_i$. Generally, a mixture model is written in the form of:

\[f_{mix}(x; \Theta, \pi) = \sum_{k=1}^K \pi_k f_k(x)\]

Where $f_i(x)$ is called the ith component and $\pi_i$ is called the ith mixing coeffiecent.

Gaussian Mixture Model

StateSpaceDynamics.GaussianMixtureModel — Method
GaussianMixtureModel(k::Int, data_dim::Int)

Constructor for GaussianMixtureModel. Initializes Σₖ's covariance matrices to the identity, πₖ to a uniform distribution, and μₖ's means to zeros.

source
StateSpaceDynamics.fit! — Method
fit!(gmm::GaussianMixtureModel, data::AbstractMatrix{<:Real}; <keyword arguments>)

Fits a Gaussian Mixture Model (GMM) to the given data using the Expectation-Maximization (EM) algorithm.

Arguments

  • gmm::GaussianMixtureModel: The Gaussian Mixture Model to be fitted.
  • data::AbstractMatrix{<:Real}: The dataset on which the model will be fitted, where each row represents a data point.
  • maxiter::Int=50: The maximum number of iterations for the EM algorithm (default: 50).
  • tol::Float64=1e-3: The tolerance for convergence. The algorithm stops if the change in log-likelihood between iterations is less than this value (default: 1e-3).
  • initialize_kmeans::Bool=false: If true, initializes the means of the GMM using K-means++ initialization (default: false).

Returns

  • class_probabilities: A matrix where each entry (i, k) represents the probability of the i-th data point belonging to the k-th component of the mixture model.
source
StateSpaceDynamics.loglikelihood — Method
loglikelihood(gmm::GaussianMixtureModel, data::AbstractMatrix{<:Real})

Compute the log-likelihood of the data given the Gaussian Mixture Model (GMM). The data matrix should be of shape (# observations, # features).

Arguments

  • gmm::GaussianMixtureModel: The Gaussian Mixture Model instance
  • data::AbstractMatrix{<:Real}: data matrix to calculate the Log-Likelihood

Returns

  • Float64: The log-likelihood of the data given the model.
source

Poisson Mixture Model

StateSpaceDynamics.PoissonMixtureModel — Type
PoissonMixtureModel

A Poisson Mixture Model for clustering and density estimation.

Fields

  • k::Int: Number of poisson-distributed clusters.
  • λₖ::Vector{Float64}: Means of each cluster.
  • πₖ::Vector{Float64}: Mixing coefficients for each cluster.
source
StateSpaceDynamics.fit! — Method
fit!(pmm::PoissonMixtureModel, data::AbstractMatrix{<:Integer}; maxiter::Int=50, tol::Float64=1e-3, initialize_kmeans::Bool=false)

Fits a Poisson Mixture Model (PMM) to the given data using the Expectation-Maximization (EM) algorithm.

Arguments

  • pmm::PoissonMixtureModel: The Poisson Mixture Model to be fitted.
  • data::AbstractMatrix{<:Integer}: The dataset on which the model will be fitted, where each row represents a data point.

Keyword Arguments

  • maxiter::Int=50: The maximum number of iterations for the EM algorithm (default: 50).
  • tol::Float64=1e-3: The tolerance for convergence. The algorithm stops if the change in log-likelihood between iterations is less than this value (default: 1e-3).
  • initialize_kmeans::Bool=false: If true, initializes the means of the PMM using K-means++ initialization (default: false).

Returns

  • class_probabilities: A matrix where each entry (i, k) represents the probability of the i-th data point belonging to the k-th component of the mixture model.
source
StateSpaceDynamics.loglikelihood — Method
loglikelihood(pmm::PoissonMixtureModel, data::AbstractMatrix{<:Integer})

Compute the log-likelihood of the data given the Poisson Mixture Model (PMM). The data matrix should be of shape (# features, # obs).

Returns

  • Float64: The log-likelihood of the data given the model.
source