Keep *_model = NULL to proceed with a model made from default parameters (see each individual make_* function for parameters). However, it is not advisable to accept default parameters for each model; you should tune each model according to your dataset and the business question you are answering.
Usage
bt_compile_model(
...,
embedding_model = NULL,
reduction_model = NULL,
clustering_model = NULL,
vectoriser_model = NULL,
ctfidf_model = NULL
)Arguments
- ...
Additional arguments sent to bertopic.BERTopic()
- embedding_model
Model for creating embeddings (Python object)
- reduction_model
Model for reducing embeddings' dimensions (Python object)
- clustering_model
Model for clustering (Python object)
- vectoriser_model
Model for vectorising input for topic representations (Python object)
- ctfidf_model
Model for performing class-based tf-idf (ctf-idf) (Python object)
Examples
if (FALSE) {
# model using all default parameters
model <- bt_compile_model()
# model with modular components already generated
# define embedding and reduction modules and pass to bt_compile_model
embedder <- bt_make_embedder_st("all-mpnet-base-v2")
reducer <- bt_make_reducer_umap(n_components = 10L, n_neighbours = 20L)
model <- bt_compile_model(embedding_model = embedder, reduction_model = reducer)
# Perform document embedding and reduction external to bertopic model and pass empty models to bt_compile_model
embedder <- bt_make_embedder_st("all-mpnet-base-v2") # embedder
embeddings <- bt_do_embedding(embedder, docs, accelerator = NULL) # embeddings
reducer <- bt_make_reducer_umap(n_components = 10L, n_neighbours = 20L) # reducer
reduced_embeddings <- bt_do_reducing(reducer, embeddings) # reduced embeddings
# skip embedding and reduction step by passing empty models
model <- bt_compile_model(embedding_model = bt_empty_embedder(), reduction_model = bt_empty_reducer())
}