🚧

.github/workflows/docker-gpu-image.yml

@@ -0,0 +1,39 @@
+name: GPU Image
+on:
+  workflow_dispatch: null
+  push:
+    paths: 
+      - Dockerfile.gpu
+      - rl-env.yml
+      - spatial-env.yml
+      - jupyter-ai.yml
+      - install.r
+jobs:
+  build:
+    runs-on: ubuntu-latest
+    permissions: write-all
+    steps:
+    # For biggish images, github actions runs out of disk space.
+    # So we cleanup some unwanted things in the disk image, and reclaim that space for our docker use
+    # https://github.com/actions/virtual-environments/issues/2606#issuecomment-772683150
+    # and https://github.com/easimon/maximize-build-space/blob/b4d02c14493a9653fe7af06cc89ca5298071c66e/action.yml#L104
+    # This gives us a total of about 52G of free space, which should be enough for now
+      - name: cleanup disk space
+        run: |
+          sudo rm -rf /usr/local/lib/android /usr/share/dotnet /opt/ghc
+          df -h
+      - uses: actions/checkout@v3
+      - name: Login to GitHub Container Registry
+        if: github.repository == 'rocker-org/ml'
+        uses: docker/login-action@v1
+        with:
+          registry: ghcr.io
+          username: ${{github.actor}}
+          password: ${{secrets.GITHUB_TOKEN}}
+      - name: Build the Docker image
+        if: github.repository == 'rocker-org/ml'
+        run: docker build images/ --no-cache -f images/Dockerfile.cuda --tag ghcr.io/rocker/cuda:latest
+      - name: Publish
+        if: github.repository == 'rocker-org/ml'
+        run: docker push ghcr.io/rocker/cuda:latest
+

.github/workflows/docker-image.yml

@@ -0,0 +1,34 @@
+name: Docker Image CI
+on:
+  workflow_dispatch: null
+  push:
+    paths: ['images/Dockerfile',  'images/install.r', 'images/rl-env.yml', 'images/spatial-env.yml', 'images/jupyter-ai.yml']
+jobs:
+  build:
+    runs-on: ubuntu-latest
+    permissions: write-all
+    steps:
+    # For biggish images, github actions runs out of disk space.
+    # So we cleanup some unwanted things in the disk image, and reclaim that space for our docker use
+    # https://github.com/actions/virtual-environments/issues/2606#issuecomment-772683150
+    # and https://github.com/easimon/maximize-build-space/blob/b4d02c14493a9653fe7af06cc89ca5298071c66e/action.yml#L104
+    # This gives us a total of about 52G of free space, which should be enough for now
+      - name: cleanup disk space
+        run: |
+          sudo rm -rf /usr/local/lib/android /usr/share/dotnet /opt/ghc
+          df -h
+      - uses: actions/checkout@v3
+      - name: Login to GitHub Container Registry
+        if: github.repository == 'boettiger-lab/k8s'
+        uses: docker/login-action@v1
+        with:
+          registry: ghcr.io
+          username: ${{github.actor}}
+          password: ${{secrets.GITHUB_TOKEN}}
+      - name: Build the Docker image
+        if: github.repository == 'boettiger-lab/k8s'
+        run: docker build images/ --tag ghcr.io/boettiger-lab/k8s:latest
+      - name: Publish
+        if: github.repository == 'boettiger-lab/k8s'
+        run: docker push ghcr.io/boettiger-lab/k8s:latest
+

Dockerfile

@@ -0,0 +1,35 @@
+ARG BASE=quay.io/jupyter/minimal-notebook:ubuntu-24.04
+FROM $BASE
+
+# Make code-server extensions etc persist to container, not hub
+ENV XDG_DATA_HOME=/opt/share
+
+USER root
+# code-server (VSCode)
+RUN curl -fsSL https://code-server.dev/install.sh | sh && rm -rf .cache 
+
+# apt utilities, code-server setup
+RUN curl -s https://raw.githubusercontent.com/rocker-org/ml/refs/heads/main/install_utilities.sh | bash
+
+RUN curl -s https://raw.githubusercontent.com/rocker-org/ml/refs/heads/main/install_r.sh | bash
+RUN curl -s https://raw.githubusercontent.com/rocker-org/ml/refs/heads/main/install_rstudio.sh | bash
+
+## Add rstudio's binaries to path for quarto
+ENV PATH=$PATH:/usr/lib/rstudio-server/bin/quarto/bin
+
+## switch non-root users from BSPM to r-universe if no sudo 
+COPY Rprofile /usr/lib/R/etc/Rprofile.site
+
+# When run as root, install.r automagically handles any necessary apt-gets
+COPY install.r install.r
+RUN Rscript install.r
+
+USER ${NB_USER}
+
+COPY vscode-extensions.txt vscode-extensions.txt
+RUN xargs -n 1 code-server --install-extension < vscode-extensions.txt
+
+COPY environment.yml environment.yml
+RUN conda update --all --solver=classic -n base -c conda-forge conda && \
+    conda env update --file environment.yml
+

Dockerfile.cuda

@@ -0,0 +1,35 @@
+ARG BASE=quay.io/jupyter/pytorch-notebook:cuda12-ubuntu-24.04
+FROM $BASE
+
+# Make code-server extensions etc persist to container, not hub
+ENV XDG_DATA_HOME=/opt/share
+
+USER root
+# code-server (VSCode)
+RUN curl -fsSL https://code-server.dev/install.sh | sh && rm -rf .cache 
+
+# apt utilities, code-server setup
+RUN curl -s https://raw.githubusercontent.com/rocker-org/ml/refs/heads/main/install_utilities.sh | bash
+
+RUN curl -s https://raw.githubusercontent.com/rocker-org/ml/refs/heads/main/install_r.sh | bash
+RUN curl -s https://raw.githubusercontent.com/rocker-org/ml/refs/heads/main/install_rstudio.sh | bash
+
+## Add rstudio's binaries to path for quarto
+ENV PATH=$PATH:/usr/lib/rstudio-server/bin/quarto/bin
+
+## switch non-root users from BSPM to r-universe if no sudo 
+COPY Rprofile /usr/lib/R/etc/Rprofile.site
+
+# When run as root, install.r automagically handles any necessary apt-gets
+COPY install.r install.r
+RUN Rscript install.r
+
+USER ${NB_USER}
+
+COPY vscode-extensions.txt vscode-extensions.txt
+RUN xargs -n 1 code-server --install-extension < vscode-extensions.txt
+
+COPY environment.yml environment.yml
+RUN conda update --all --solver=classic -n base -c conda-forge conda && \
+    conda env update --file environment.yml
+

README.md

@@ -1,77 +1,3 @@
 # Rocker stack for Machine Learning in R 
 
-This repository contains images for machine learning and GPU-based computation in R.  **EDIT** Dockerfiles are now built in modular build system at https://github.com/rocker-org/rocker-versioned2 .  This repo remains for documentation around the ML part of the stack.  
-
-
-
-
-The dependency stack looks like so: 
-
-```
--| rocker/cuda
-  -| rocker/ml
-    -| rocker/ml-verse
-```
-
-All three are CUDA compatible and will optionally take R version tags (`rocker/ml:4.0.5`) with the option of additional trailing CUDA version tag (e.g. `rocker/ml:4.0.5-cuda10.1`). 
-
-
-## Quick start
-
-**Note: GPU use requires [nvidia-docker](https://github.com/NVIDIA/nvidia-docker/)** runtime to run!  
-
-Run a bash shell or R command line:
-
-```
-# CPU-only
-docker run --rm -ti rocker/ml R
-# Machines with nvidia-docker and GPU support
-docker run --gpus all --rm -ti rocker/ml R
-```
-
-Or run in RStudio instance:
-
-```
-docker run --gpus all -e PASSWORD=mu -p 8787:8787 rocker/ml
-```
-
-
-## Tags
-
-See [current `ml` tags](https://hub.docker.com/r/rocker/ml/tags?page=1&ordering=last_updated)
-See [current `ml-verse` tags](https://hub.docker.com/r/rocker/ml-verse/tags?page=1&ordering=last_updated)
-
-
-## Python versions and virtualenvs
-
-The ML images configure a default python virtualenv using the Ubuntu system python (3.8.5 for current Ubuntu 20.04 LTS), see [install_python.sh](https://github.com/rocker-org/rocker-versioned2/blob/master/scripts/install_python.sh).  This virtualenv is user-writable and the default detected by `reticulate` (using `WORKON_HOME` and `PYTHON_VENV_PATH` variables).
-
-Images also configure [pipenv](https://github.com/pypa/pipenv) with [pyenv](https://github.com/pyenv/pyenv) by default.  This makes it very easy to manage projects that require specific versions of Python as well as specific python modules.  For instance, a project using the popular `[greta](https://greta-stats.org/)` package for GPU-accelerated Bayesian inference needs Tensorflow 1.x, which requires Python <= 3.7, might do:
-
-```bash
-pipenv --python 3.7
-```
-
-In the bash terminal to set up a pipenv-managed virtualenv in the working directory using Python 3.7.  Then in R we can activate this virtualenv
-
-```r
-venv <- system("pipenv --venv", inter = TRUE)
-reticulate::use_virtualenv(venv, required = TRUE)
-```
-
-We can now install tensorflow version needed, e.g.
-
-```r
-install.packages("tensorflow")
-tensorflow::install_tensorflow(version="1.14.0-gpu", extra_packages="tensorflow-probability==0.7.0")
-```
-
-
-
-## Notes
-
-All images are based on the current Ubuntu LTS (ubuntu 20.04) and based on the official [NVIDIA CUDA docker build recipes](https://gitlab.com/nvidia/container-images/cuda/)
-
-**PLEASE NOTE**: older images, `rocker/ml-gpu`, `rocker/tensorflow` and `rocker/tensorflow-gpu`, built with cuda 9.0, are deprecated and no longer supported.  
-
-
+This repository contains images for machine learning and GPU-based computation in R.  

Rprofile

@@ -0,0 +1,42 @@
+# Based on Jeroen's config: 
+# https://github.com/r-universe-org/base-image/blob/f20ec9fc6f51ef8a89aad489206a43790bd9bf77/Rprofile
+
+
+
+local({
+
+has_sudo <- function() {
+  system("sudo -n true", intern = TRUE) == 0
+}
+
+# Only use Runiverse if we don't have sudo
+if (!has_sudo()) {
+
+  rver <- getRversion()
+  distro <- system2('lsb_release', '-sc', stdout = TRUE)
+  options(HTTPUserAgent = sprintf("R/%s R (%s)", rver, paste(rver, R.version$platform, R.version$arch, R.version$os)))
+  options(repos = c(CRAN = sprintf("https://packagemanager.rstudio.com/all/__linux__/%s/latest", distro)))
+
+  # Enable BioConductor repos
+  utils::setRepositories(ind = 1:4, addURLs = c(fallback = "https://cloud.r-project.org"))
+
+  # Enable universe repo(s)
+  my_universe <- Sys.getenv("MY_UNIVERSE", "https://cran.r-universe.dev")
+  if(nchar(my_universe)){
+    my_repos <- trimws(strsplit(my_universe, ';')[[1]])
+    binaries <- sprintf('%s/bin/linux/%s/%s', my_repos[1], distro, substr(rver, 1, 3))
+    options(repos = c(binaries = binaries, universe = my_repos, getOption("repos")))
+  }
+
+}
+# Other settings
+options(crayon.enabled = TRUE)
+Sys.unsetenv(c("CI", "GITHUB_ACTIONS"))
+
+# Dummy token for API limits
+if(is.na(Sys.getenv("GITHUB_PAT", NA))){
+  dummy <- c('ghp_SXg', 'LNM', 'Tu4cnal', 'tdqkZtBojc3s563G', 'iqv')
+  Sys.setenv(GITHUB_PAT = paste(dummy, collapse = 'e'))
+}
+})
+