{wrappingtransformers}

{wrappingtransformers} provides acess to some functions of some pretrained NLP models shared on Hugging Face’s model community platform through R that you’d normally access using Python.

• Currently implemented:
  • flair/ner-german-large
    • sequence tagger
• Planned:
  • flair/ner-german
    • sequence tagger
  • deepset/bert-base-german-cased-hatespeech-GermEval18Coarse
    • text classification

Basic Installation

Because {wrappingtransformers} has Python dependencies, the setup is a bit more complicated than usual — you’ll have to run some extra commands before you can use it.

If you just want to use transformers from within R, run the following commands in R. This will create a local Python installation that will only be used by R and install the needed Python package for you.

install.packages("reticulate")
reticulate::install_miniconda(update = TRUE) # Install Python
reticulate::py_install('transformers') # Install the Python transformers package
remotes::install_github("balthasars/wrappingtransformers") # Install this package

After running these commands restart R.

Advanced Setup — Select the Python version to be used

Discover what version of Python will be used unless you set it manually as suggested below. transformers is tested with Python 3.6. In my experience >= 3.9 works best.

reticulate::py_discover_config()

If you’re happy with what this indicates, you may skip over the following steps.

If not, set the environmental varible RETICULATE_PYTHON to the desired Python version path interactively with usethis::edit_r_environ() or using Sys.setenv():

Sys.setenv(RETICULATE_PYTHON = "/usr/local/bin/python3.9")

Alternatively you can select the Python environment using the reticulate package:

reticulate::use_python(python = "/usr/local/bin/python3.9")

Select Environment

Use create_wrappingtransformers_venv() to create a new environment solely for the purpose of using this package.

This will install all required Python libraries using the version of python indicated in the environment variable RETICULATE_PYTHON if you don’t indicate it in argument python_path.

create_wrappingtransformers_venv(python_path = NULL)

If you already have a Python environment with transformers installed, name it in the virtualenv argument. This will set it up for the rest of this R session.

reticulate::use_virtualenv(virtualenv = "myenv")

or use Conda:

reticulate::use_condaenv(condaenv = "myenv")

Using the models

flair/german-ner-large

Here’s a full example on how to use the NER tagger:

reticulate::use_python(python = "/usr/local/bin/python3.9", required = TRUE)
reticulate::use_virtualenv("wrappingtransformers", required = TRUE)
wt_load_model(model_name = "ner-german-large")
sentence_example <- "Santesuisse und Curaviva fordern eine Umverteilung"
wt_predict_labels_ner_german_large(sentence = sentence_example)

deepset/bert-base-german-cased-hatespeech-GermEval18Coarse

Here’s a full example on how to perform sequence classification:

reticulate::use_python(python = "/usr/local/bin/python3.9", required = TRUE)
reticulate::use_virtualenv(virtualenv = "wrappingtransformers", required = TRUE)
load_bert_base_german_cased_hatespeech()
sentence_example <- "Der Wichser muss raus."
wt_predict_bert_base_german_cased_hatespeech(sentence = sentence_example)