data_gen: data_gen

In cvraut/viLDA: variational inference Latent Dirichlet Allocation

Description

Function to generate a simulated dataset following the LDA model.

Arguments

n_doc	The number of documents to generate
n_vocab	The number of words in the corpus
n_top	The number of topics/clusters (K)
doc_length_scale	A number proportional to the average number of words in a document (default = 8)
doc_length_scale_var	A number proportional to the variance of the average number of words in a document (default = 2)
voc_p_scale	A number proportional to the initial probability of each word in a cluster. (default = 4) The higher, the less uniform weight gets applied across all topics.
spike_overlap	A number proportional to the amount of vocabulary shared across documents from different clusters. (default = 0.05) The default value of 0.05 means that documents from different clusters will share ~5% of their word distributions with each other.
alphaWords	Hyperparameter for document-cluster distribution (default = 0.2)
alphaTopics	Hyperparameter for topic-cluster distribution (default = 0.2)
seed	The random seed for the data generation (ran once at beginning of function, default = 19890418)
topic_mix	Boolean flag, if TRUE then each document can be generated from different topic clusters (default = FALSE)
DEBUG	Boolean flag, if TRUE then debug print statements are shown to the user (default = FALSE)

Value

list("dat","word_dist","gen_topics","doc_len")

• $dat: dataframe of the document_id-word_id-count data

• $word_dist: matrix of the word-topic distributions

• $gen_topics: the selected topic for each document

• $doc_len: a n_doc length vector of the number of words in each document

cvraut/viLDA documentation built on Dec. 19, 2021, 7:05 p.m.