NEWS

---

MagmaClustR 1.2.1 (2024-06-28)

Major

• Add a precondition_hp() function and a whole pipeline to provide data_driven initialisation and pre-explore the HP space before actual MT/MO GP training.
• Add support for multiple latent processes in the multi-output convolution kernel (convolution_kernel()): the covariance is now a sum over Q latent processes, enabling richer (rank-Q) cross-output correlation structures. Initialise the corresponding hyper-parameters with hp(..., n_latent = Q).
• Change the way prior means work in the MagmaClust framework. They are now optimised during Mstep using a closed-form optimum.
• Update the simu_db() (and add an alias simu_data()) function to handle multi-output scenarios.
• Update the plot_db() function to handle multi-output scenarios, and 2D inputs.
• Add the 'train_shared_gp()' function to provide more robust initialisation of hyperparameters.

Minor

• Fix the deprecation issues of tidyselect() through .data$
• The model assumption 'common_hp' are now named 'shared_hp' instead for clarity.
• The mean process is simulated via generate_mean_process(), whereas task data are derived from generate_single_task_data(). simu_indiv_se() has been removed from the package.
• Add the 'plot_clusters()' function to display clustering results directly after training
• Add an argument to display testing data directly within plotting functions
• Fix the plot_gif() function to avoid saving multiple temporary png files
• Add a minimal implementation of Laplace Matching for binary classification
• Fix a bug when plotting in 2D mean process predictions
• Fix a few bugs when using pred_*() functions with argument data = NULL
• Simplify and improve visualisation with plot_db() and export it for users
• Replace the deprecated aes_string function with modern syntax of aes
• Fix the wrong definition of GPs mixtures by drawing a cluster instead of averaging
• Fix a bug when using pred_magmaclust() with no data argument.
• Add convolution_kernel() function to generate the convolution process kernel
• kern_to_cov() now natively supports multi-output scenarios. The convolution kernel has been added to the list of documented and supported kernels.
• Added a safety check before returning the covariance matrix. If the matrix contains NaN or Inf values, an explicit warning is now triggered.
• Update hp() to natively support multi-output scenarios.
• Add extensive logic to initialize HPs specifically for the convolution kernel.
• Modernize the internal logic for building the HPs tables in hp().
• In hp(), the chained if/else logic of standard string kernels has been replaced by a 'switch' statement.
• In hp(), 'list_hp' argument has been removed, as custom parameter generation is now safely handled via the 'hp_config' workflow.
• Add format_to_legacy() to reformat from new to old conventions for datasets' column names.

Major

• Make 'get_full_cov = T' the default option when using pred_magmaclust()

Minor

• Remove the useless messages when using samples=T in plotting functions
• Create an error when users try train_magmaclust(nb_cluster = 1)
• Make 'samples = T' the default plot when using prediction functions
• Add mention of the 'hp()' function to the documentation
• Fix Warning message when plotting clusters with exact same posterior probability
• Display plots for the generic (no data provided) prediction that are condistent with classic ones
• Disable the symmetry check when generating posterior samples for predictions

MagmaClustR 1.2.0 (2024-03-21)

Major

• Add new functions sample_gp() and sample_magmaclust() to sample from the posterior predictions of GPs, Magma, and MagmaClust
• Propose a new visualisation based on posterior samples instead of Credible Intervals for both plot_gp() and plot_magmaclust()
• Allow pred_*() functions to be used without providing the 'data' argument by returning the mean processes (or priors) as generic predictions.

Minor

• Add an option to generate multiple curves from sample_gp()
• Fix a bug when the 'Reference' column is present when using plot_gp()
• Fix the '\docType{package}' bug in roxygen2
• Add 'plot_magma()' as a duplicated name for 'plot_gp()'
• Remove the useless 'Reference' column in the prediction of the mean processes
• Fix a bug when both 'pred' and 'samples' are provided in plot_samples()

MagmaClustR 1.1.2 (2023-05-23)

Minor

• Fix a bug occurring in pred_magmaclust() for a 'trained_model' with hp_i=FALSE
• Simplify the use of hyperposterior() and hyperposterior_clust() by providing a 'trained_model' argument.

MagmaClustR 1.1.1 (2023-01-17)

Minor

• Fix an issue regarding deprecation of .data in 'tidyselect'
• Fix a unit testing issue

MagmaClustR 1.1.0 (2022-11-01)

Major

• Provide 4 vignettes explaining in detail how the different features of MagmaClustR work on practical examples.
• Implement expand_grid_inputs() to help create customised n-dimensional input grids on which to evaluate the GP.
• Implement regularize_data() to project a dataset on a specific input grid, (possibly to control the size of the resulting covariance matrices and the associated running time).
• Add an internal 'Reference' column to datasets, to provide an adequate identifier for multidimensional inputs.
• Implement a new version of simu_db() to generate more realistic 2-D datasets.

Minor

• Round inputs to 6 significant digits to avoid numerical errors.
• Generalise the creation of a grid in any dimension when 'grid_inputs' is not specified in the prediction functions.

MagmaClustR 1.0.1 (2022-08-29)

Major

*Remove the package 'optimr' dependency and switch to base 'optim()' function *Increase convergence tolerance in 'optim()', which was too slow

Minor

*Fix the warnings about the absolute value function in the Cpp code *Remove error message in 'train_magmaclust()' when common_hp_k = FALSE *Change the default intervals for hyper-parameters in 'simu_db()' *Automatically remove rows with missing data *Change position of the 'grid_inputs' argument in prediction functions *Remove the internal functions from the index documentation *Fix 'ID' in hyperposterior() and hyperposterior_clust() when not character

MagmaClustR 1.0.0 (2022-06-06)

Initial release