Using spacyr with Docker

A Docker image can contain R, spacyr, Python, spaCy and the language model. Installing them at build time lets you start a container and parse text straight away. The Dockerfile below uses the Rocker image for R and offers two build targets. The manual target installs a specific spaCy version and model with pip. The managed target uses spacy_install() to create the Python environment and install the default English model.

Prepare the build directory

With Docker running, open a terminal in a Git checkout of spacyr. Create a build directory and export the checked-out revision into it:

build_dir=$(mktemp -d)
git archive --output="$build_dir/spacyr-src.tar" HEAD
cd "$build_dir"

Save the following as Dockerfile in that directory, next to spacyr-src.tar:

FROM rocker/r-ver:4.5.1 AS base

RUN apt-get update \
    && apt-get install -y --no-install-recommends libpng-dev libpython3-dev python3-venv \
    && rm -rf /var/lib/apt/lists/*

RUN Rscript --vanilla -e \
    'install.packages(c("data.table", "reticulate"), repos = "https://cloud.r-project.org", Ncpus = 2L)'
ADD spacyr-src.tar /tmp/spacyr/
RUN R CMD INSTALL /tmp/spacyr && rm -rf /tmp/spacyr

FROM base AS manual

RUN python3 -m venv /opt/spacy \
    && /opt/spacy/bin/python -m pip install --no-cache-dir \
        'spacy==3.8.18' \
        'https://github.com/explosion/spacy-models/releases/download/en_core_web_sm-3.8.0/en_core_web_sm-3.8.0-py3-none-any.whl'
ENV RETICULATE_PYTHON=/opt/spacy/bin/python

FROM base AS managed

RUN apt-get update \
    && apt-get install -y --no-install-recommends python3-pip \
    && rm -rf /var/lib/apt/lists/*

ENV SPACY_PYTHON=/opt/spacy
RUN Rscript --vanilla -e 'spacyr::spacy_install(ask = FALSE)'

Build with pinned spaCy and model versions

docker build --target manual --tag spacyr:manual .

The manual target installs spaCy 3.8.18 and the English pipeline en_core_web_sm 3.8.0 in /opt/spacy. The model wheel is installed from its release on GitHub. Set RETICULATE_PYTHON to the Python executable in the virtual environment so reticulate uses it when spacy_initialize() starts Python.

Build with spacy_install()

docker build --target managed --tag spacyr:managed .

The managed target installs spaCy and the default English model during the build. Use ask = FALSE to make the installation non-interactive. This is already the default under Rscript. On this Rocker image, reticulate needs both python3-venv and python3-pip to create the virtual environment from system Python.

Set SPACY_PYTHON to the environment directory so the build and subsequent R sessions use the same location, even under different user accounts. Leave RETICULATE_PYTHON unset for this target. To choose a spaCy release or add other models, pass version or lang_models to spacy_install().

Check the finished image

Run a parse as an unprivileged user with networking disabled. To check the managed installation, use spacyr:managed in the same command.

docker run --rm -i --network none --user 10001:10001 --env HOME=/tmp \
    spacyr:manual Rscript --vanilla - <<'RSCRIPT'
spacyr::spacy_initialize(model = "en_core_web_sm")
parsed <- spacyr::spacy_parse("Docker works.")
stopifnot(identical(as.character(parsed$token), c("Docker", "works", ".")))
print(parsed)
RSCRIPT

Both targets were tested with Docker Desktop on Apple silicon (Linux ARM64).