Dockerfile for generic ML environment

FROM nvidia/cuda:10.2-base

RUN apt-get update && \
    apt-get install -y --no-install-recommends git

SHELL ["/bin/bash", "-c"]

#setup miniconda
ENV CONDAROOT "/opt/conda"

WORKDIR /root/
ADD https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh /root/

RUN mkdir ~/.conda && \
    bash Miniconda3-latest-Linux-x86_64.sh -b -p $CONDAROOT && \
    #rm -rf Miniconda3-latest-Linux-x86_64.sh && \
    ln -s $CONDAROOT/etc/profile.d/conda.sh /etc/profile.d/conda.sh && \
    source $CONDAROOT/etc/profile.d/conda.sh

# add conda to path
ENV PATH $CONDAROOT/bin:$PATH

# install conda packages, -c are channels for dependencies, -n are virtual env
RUN conda update -n base -c defaults conda && \
    conda create -n torch -y python=3.7 && \
    conda install -n torch pytorch torchvision torchaudio cudatoolkit=10.2 -c pytorch && \
    conda install -n torch -c fastai -c pytorch fastai && \
    conda install -n torch -c conda-forge imageio matplotlib seaborn pandas jupyter jupyterlab scikit-image scikit-learn tqdm jupyter_contrib_nbextensions tensorboard grpcio && \
#explicit nodejs version needed
    conda install -n torch -c conda-forge nodejs=10.13 && \
#issue:
#python graphviz has errors if you don't include certain X server libraries not marked as dependencies
#error is "cannot find libgvplugin_pango.so.6 or one of its dependencies try ldd"
#goto directory with libgvplugin_pango.so.6 and ldd the file, ensure no dependencies are missing
    conda install -n torch -c conda-forge python-graphviz xorg-libxrender xorg-libxpm xorg-libxau

# see about adding optimized pillow at some later time

# activate torch environment to install packages that are outdated or not available through conda
ENV PATH $CONDAROOT/envs/torch/bin:$PATH
RUN echo "source activate torch" >> ~/.bashrc && \
    source activate torch && \
    pip install opencv-python albumentations pretrainedmodels efficientnet-pytorch torchsummary future absl-py jupyter-tensorboard hiddenlayer && \
    pip install --no-dependencies git+https://github.com/qubvel/segmentation_models.pytorch

#configure jupyter-lab to run in the docker image as root with bash as terminal and no password
# notebook directory is /opt/notebooks ==> this should be your mount point
RUN jupyter-lab --generate-config

RUN sed -i '/c.NotebookApp.notebook_dir/c\c.NotebookApp.notebook_dir = "'"/opt/notebooks"'"' ~/.jupyter/jupyter_notebook_config.py && \
    sed -i '/c.NotebookApp.open_browser/c\c.NotebookApp.open_browser = False' ~/.jupyter/jupyter_notebook_config.py && \
    sed -i '/c.NotebookApp.quit_button/c\c.NotebookApp.quit_button = True' ~/.jupyter/jupyter_notebook_config.py && \
    sed -i '/c.NotebookApp.token/c\c.NotebookApp.token = "'""'"' ~/.jupyter/jupyter_notebook_config.py && \
    sed -i '/c.NotebookApp.ip/c\c.NotebookApp.ip = "'"0.0.0.0"'"' ~/.jupyter/jupyter_notebook_config.py && \
    sed -i '/c.NotebookApp.terminado_settings/c\c.NotebookApp.terminado_settings = {"'"shell_command"'":["'"bash"'"]}' ~/.jupyter/jupyter_notebook_config.py && \
    sed -i '/c.NotebookApp.allow_root/c\c.NotebookApp.allow_root = True' ~/.jupyter/jupyter_notebook_config.py && \
    jupyter labextension install @jupyter-widgets/jupyterlab-manager && \
    jupyter labextension install jupyterlab_tensorboard && \
    mkdir /opt/notebooks
WORKDIR /opt/notebooks

# Set the random seed and copy the utility scripts to the image
ENV RANDOM_SEED 2019
COPY torchtest.py seed.py /opt/scripts/

# this script runs seed.py whenever an ipython kernel/console is started
# this actually creates a dir in /opt/notebooks/'~'/.ipython
COPY ipython_config.py ~/.ipython/profile_default/

RUN mkdir /opt/cache && \
    pip freeze > ~/requirements.txt && \
    conda list -n torch --export --json > ~/requirements.json
ENV TORCH_HOME ~/cache/torch
ENV FASTAI_HOME ~/cache/fastai
ENV HOME /root/

RUN conda install -c fastai -c pytorch fastbook

# Make port 8888 available to the world outside this container [doesn't work, use docker run option -p 8888:8888]
EXPOSE 8888

# Run the torchtest script when the container launches (and no other command is given)
#CMD ["ipython", "/opt/scripts/torchtest.py"]
CMD ["/bin/bash", "-c", "jupyter lab"]