MixtureDensityNetworks v0.3.0

This release adds support for Flux 0.14 and drops support for Julia < 1.9.

Features

• Added support for Flux 0.14.
• Dropped support for Julia < 1.9.
• Progress bars now display properly depending on whether code is running in the terminal or in Pluto.

MixtureDensityNetworks v0.2.2

This release replaces the full covariance matrix in the MultivariateGMM layer with a diagonal covariance matrix.

Features

• This version replaces the full covariance matrix in the MultivariateGMM layer with a diagonal covariance matrix. This provides significantly faster training time at the cost of representation power. However, we can regain much of this power by increasing the number of mixture components in the predicted distribution.

MixtureDensityNetworks v0.2.1

This release fixes a bug where the identity matrix was undefined.

Features

• Fixed bug where Identity matrix was undefined.

MixtureDensityNetworks v0.2.0

This release introduces support for multivariate regression with the new MultivariateGMM layer.

Features

• Introduces support for multivariate regression with the new MultivariateGMM layer.
• MultivariateGMM maintains the full covariance matrix for each Gaussian Mixture, which enables the modelling of relations between labels.
• The new and UnivariateGMM and MultivariateGMM layers allow users to add univariate and multivariate GMMs to pre-existing models.
• The native interface was reworked, which introduces breaking changes for projects relying on it. The MLJ interface remains unchanged.

MixtureDensityNetworks v0.1.0

This release introduces full compatibility with MLJ for training, evaluating, and predicting models.

Features

Implements the MLJModelInterface.

MixtureDensityNetworks v0.0.2

This is the first official release of MixtureDensityNetworks.

Features

Native API
Documentation