do_pca: Do PCA on Embeddings

In jonathanbratt/RBERTviz: Visualization Tools for RBERT

Description

Calculate the 2D PCA for a collection of embedding vectors, in preparation for plotting.

Usage

do_pca(
  embedding_df,
  project_vectors = embedding_df,
  disambiguate_tokens = TRUE
)

Arguments

embedding_df	A tbl_df of embedding vectors; from the output of extract_features.
project_vectors	A tbl_df of embedding vectors to be used for calculating the PCA projection matrix. Defaults to embedding_df. This makes it possible to more consistently select the PCA "perspective", even as the set of vectors may change.
disambiguate_tokens	Logical; whether to append example and token index to the literal token for display purposes.

Value

A tbl_df of the embedding vectors projected onto two principal axes.

Examples

## Not run: 
# assuming something like the following has been run:
# feats <- RBERT::extract_features(...) # See RBERT documentation
# Then:
pca_df <- feats$output %>%
    filter_layer_embeddings(layer_indices = 12L) %>%
    keep_tokens("[CLS]") %>%
    do_pca()

## End(Not run)

jonathanbratt/RBERTviz documentation built on June 19, 2021, 6:27 p.m.