In farach/huggingfaceR: Hugging Face State-of-the-Art Models in R

knitr::opts_chunk$set(
  collapse = TRUE,
  comment = "#>",
  fig.path = "man/figures/README-",
  out.width = "100%"
)

huggingfaceR

The goal of huggingfaceR is to to bring state-of-the-art NLP models to R. huggingfaceR is built on top of Hugging Face's transformers library; and has support for navigating the Hugging Face Hub The Hub.

Installation

Prior to installing huggingfaceR please be sure to have your python environment set up correctly.

install.packages("reticulate")
library(reticulate)

install_miniconda()

If you are having issues, more detailed instructions on how to install and configure python can be found here.

After that you can install the development version of huggingfaceR from GitHub with:

# install.packages("devtools")
devtools::install_github("farach/huggingfaceR")

Example

huggingfaceR makes use of the transformers pipline() abstraction to quickly make pre-trained language models available for use in R. In this example we will load the distilbert-base-uncased-finetuned-sst-2-english model and its tokenizer into a pipeline object to obtain sentiment scores.

library(huggingfaceR)

distilBERT <- hf_load_pipeline(
  model_id = "distilbert-base-uncased-finetuned-sst-2-english", 
  task = "text-classification"
  )

distilBERT

With the pipeline now loaded, we can begin using the model.

distilBERT("I like you. I love you")

We can use this pipeline in a typical tidyverse processing chunk. First we load the tidyverse.

library(tidyverse)

We can use the huggingfaceR hf_load_dataset() function to pull in the emotion Hugging Face dataset. This dataset contains English Twitter messages with six basic emotions: anger, fear, love, sadness, and surprise. We are interested in how well the Distilbert model classifies these emotions as either a positive or a negative sentiment.

emo <- hf_load_dataset(
  dataset = "emo", 
  split = "train", 
  as_tibble = TRUE, 
  label_name = "int2str"
  )

emo_model <- emo %>%
  sample_n(100) %>% 
  transmute(
    text,
    emotion_id = label,
    emotion_name = label_name,
    distilBERT_sent = distilBERT(text)
  ) %>%
  unnest_wider(distilBERT_sent)

glimpse(emo_model)

We can use ggplot2 to visualize the results.

emo_model |>
  mutate(
    label = paste0("Distilbert class:\n", label),
    emotion_name = str_to_title(emotion_name)
  ) |>
  ggplot(aes(x = emotion_name, y = score, color = label)) +
  geom_boxplot(show.legend = FALSE, outlier.alpha = 0.4, ) +
  scale_color_manual(values = c("#D55E00", "#6699CC")) +
  facet_wrap(~ label) +
  labs(
    title = "Reviewing Distilbert classification predictions",
    x = "Original label",
    y = "Model score",
    caption = "source:\nhttps://huggingface.co/datasets/emo"
  ) +
  theme_minimal() +
  theme(
    plot.title = element_text(hjust = 0.5),
    axis.text.x = element_text(angle = 45),
    axis.title.y = element_text(margin = margin(t = 0, r = 10, b = 0, l = 0)),
    axis.title.x = element_text(margin = margin(t = 10, r = 0, b = 0, l = 0))
  )