Create an embedding model using a model available from the Flair Library

Usage

bt_make_embedder_flair(
  model,
  ...,
  flair_class = c("FlairEmbeddings", "TransformerWordEmbeddings",
    "TransformerDocumentEmbeddings", "WordEmbeddings")
)

Arguments

model: name of model used to create embeddings
...: Additional arguments to be passed to the selected Flair class
flair_class: BertopicR is currently compatible with 4 Flair embedding classes: FlairEmbeddings, WordEmbeddings, TransformerWordEmbeddings and TransformerDocumentEmbeddings. If you chose to perform word embeddings rather than document embeddings, bt_do_embedding will pool the word embeddings for each document to calculate a mean value.

Value

an embedding model, formed according to the model defined, that can be input to be_do_embedding to create document embeddings

Examples

# \donttest{
# Flair Embedding, reducing chars_per_chunk to help with memory issues
embedder <- bt_make_embedder_flair(model = "news-forward", flair_class = "FlairEmbeddings", chars_per_chunk = 400L)
#> Error in bt_make_embedder_flair(model = "news-forward", flair_class = "FlairEmbeddings",     chars_per_chunk = 400L): flair is not in installed packages of current environment, run reticulate::py_install("flair").
#> 
#>             Note that if you receive a module not found error, you may need to instead run reticulate::py_install("flair", pip = TRUE) to force installation with pip instead of conda.

# Transformer Document Embedding
embedder <- bt_make_embedder_flair(model = "roberta-base", flair_class = "TransformerDocumentEmbeddings")
#> Error in bt_make_embedder_flair(model = "roberta-base", flair_class = "TransformerDocumentEmbeddings"): flair is not in installed packages of current environment, run reticulate::py_install("flair").
#> 
#>             Note that if you receive a module not found error, you may need to instead run reticulate::py_install("flair", pip = TRUE) to force installation with pip instead of conda.
# }