NEWS.md

In jpkrooney/rcorex: An Implementation of Total Correlation Explanation

rcorex 0.3.0

• Added 'discrete' marginal type to epicorex() function
• logpx_method parameter of biocorex() and epicorex() now defaults to 'pycorex'

rcorex 0.2.3

• Removed minmarg as argument from biocorex() and epicorex()
• Restored sig_ml minimal value to the estimate_parameters_gaussian() function to prevent divide by zero errors in the marginal_p_gaussian() function. However, now the minimal value is instead set to the machine floating point minimum value. As a result, biocorex() may sometimes produce negative tcs. This is an indication that the data is non-gaussian and such results should not be trusted. This might occur, for example, when binary data is fit with a gaussian marginal description.
• Minor code optimization to speed up calculating the discrete and bernoulli
• Summary methods added
• Experimental option logpx_method added. This allows use of an alternative method of calculating log_p_xi.

rcorex 0.2.2

• Minor code optimization to speed up calculating the gaussian marginals

rcorex 0.2.1

• Fixed bug in check_converged function. The TCs for each hidden node should have been summed at each iteration before checking convergence and they were not. This is fixed now which improves detection of convergence.

rcorex 0.2.0

• Removed sig_ml minimal value in estimate_parameters_gaussian() function
• Added minmarg to replace sig_ml minimal value. minmarg imposes a minimal value (should be a negative number) on individual log_marginal values. Default is -10.
• Added epicorex() function which allows the use of different marginal descriptions for different variables of a dataset. Added bernoulli marginal description for binary data (this is an optimised version of the discrete marginal description)

rcorex 0.1.0

• Added a NEWS.md file to track changes to the package
• Added biocorex(), plot.rcorex() and make_corex_tidygraph() functions

jpkrooney/rcorex documentation built on July 25, 2022, 1:37 a.m.