NEWS.md
In PolMine/biglda: Fast LDA Topic Modelling for Big Corpora

biglda v0.0.2

mallet_instance_list_store() renamed to instance_list_save() and mallet_instance_list_restore() renamed to instance_list_load().
as_LDA() outputs information on number of docs / alphabet size.
Method svmlight() moved here from package 'polmineR.misc'.
Internal, unexported function .mallet_cmd() has been removed.
mallet_load_topicmodel() will abort with an error, if no Mallet installation is found.
New functions save_document_topics() and load_document_topics() as more memory-efficient way to get gamma matrix from Java to R.
Function load_word_weights() now has argument normalize to reproduce Java/Mallet-side normalization when loading the matrix.
instance_list_load() will check whether file exists.
svmlight() reports document number at beginning of line.
gensim_ldamodel_as_LDA_Gibbs() integrated as a as_LDA() method.
Example Gensim topic model re-calculated with k = 25 (not 250).
Slot doclengths added to (S4) class LDA_Gibbs to avoid having to reload instance list for computing metrics.
as_LDA() will now fill slot doclenghts of output object.
as_LDA() has been rewritten to improve memory efficiency somewhat.

mallet_load_topicmodel() includes some sanity checks to ensure that loading topic model succeeds (report on JVM heap space, file size etc.).
When loading the package, a message on the JVM heap space is shown.
as_LDA() relies on
java class added to access model diagnostics for RTopicModel
Added function mallet_get_topic_model_diagnostics() that will extract model diagnostics from RTopicModel java class object.
Mapping the mallet-trained model on a class from the topicmodels class will now result in an object that can be loaded without requiring the presence of the biglda package (#2).
Mapping the (mallet) model on the LDA_Gibbs model will be much faster because a method available for the RTopicModel class is used. This implies a general change from the ParallelTopicModel class to the RTopicModel class.