JupyterLab for Data Science Platform

Miniforge 3 + JupyterLab 4 for Data Science + TensorFlow (with GPU support) + Torch (with GPU support)

This project provides a pre-packaged, pre-configured JupyterLab environment running on Miniconda with NVIDIA GPU support. It includes a curated set of data science packages, allowing you to start your data science projects with ease.

Quickstart

Preferred method for running the container is with docker-compose

Docker Compose

1. download docker-compose.yml or docker-compose-gpu.yml file
2. run docker-compose -f docker-compose.yml up command
3. open https://localhost:8888 in your browser

Docker

You can run the stellars-jupyterlab-ds container without docker-compose, just with docker. Commands below will create a container with persistent home folder (settings), certs mount (with https certificates for security) and workspace (your projects). Use the commands below:

IMPORTANT: make sure you have docker host running (i.e. Docker Desktop)

Without GPU support

docker run -p 8888:8888 \
 -v stellars-jupyterlab-ds_home:/home \
 -v stellars-jupyterlab-ds_workspace:/home/lab/workspace \
 -v stellars-jupyterlab-ds_certs:/mnt/certs \
 --name lab \
 --hostname lab \
 stellars/stellars-jupyterlab-ds:latest

With GPU support (NVIDIA)

docker run -p 8888:8888 \
 -v stellars-jupyterlab-ds_home:/home \
 -v stellars-jupyterlab-ds_workspace:/home/lab/workspace \
 -v stellars-jupyterlab-ds_certs:/mnt/certs \
 -e GPU_SUPPORT_ENABLED=1 \
 --gpus all \
 --name lab-gpu \
 --hostname lab-gpu \
 stellars/stellars-jupyterlab-ds:latest

... and open https://localhost:8888 in your browser

Key Features

• JupyterLab Extensions:

  • JupyterLab-Git extension
  • JupyterLab-LSP (Python) for enhanced autocompletion and code suggestions with documentation
  • Resource usage monitor

• TensorBoard:

  • TensorBoard server running on port 6006, with logs stored in /tmp/tf_logs to monitor your ML/AI model training and neural network development

For a complete list of installed packages, please refer to the packages manifest, which is frequently updated to ensure you have access to the best tools for your development needs.

About the Author

Name: Konrad Jelen (aka Stellars Henson)
Email: konrad.jelen+github@gmail.com
LinkedIn: Konrad Jelen

Entrepreneur, enterprise architect, and data science/machine learning practitioner with extensive software development and product management experience. Previously an experimental physicist with a strong background in physics, electronics, manufacturing, and science.

Installation

To use this environment, Docker must be installed on your system. JupyterLab 4 is provided as a Docker container, ensuring complete isolation from your system's software.

Docker Hub Repository: Stellars JupyterLab DS

Required Software

1. Docker Desktop - Includes the docker-compose command necessary to run the container.
2. docker-compose - Installed automatically with Docker Desktop.

Usage

1. Pull the latest container image:

   docker-compose pull

2. (Optional) Build the image locally:
   Note that building the image locally may take approximately 1.5 hours.

   docker-compose build

3. Start the container with your desired configuration:

   • Standard non-CUDA container:

     docker-compose up

   • Standard CUDA container:

     docker-compose -f docker-compose-gpu.yml up

   • Custom container configuration:

     docker-compose -f local/your-custom-docker-compose.yml up

   • Alternatively, use the provided bin/start* scripts.

4. Access JupyterLab:
   Open https://localhost:8888 in your browser.

5. Access TensorBoard:
   Open http://localhost:6006 in your browser.

Quick Configuration Tips:

• Use the CONDA_DEFAULT_ENV variable in the docker-compose.yml files to specify your default conda environment.

Default Settings

• Work Directory: ~/workspace
• Home Directory: /home/lab (contains user settings)
• JupyterLab Settings: Stored in /home/lab/.jupyter
• Volume Mapping: If using docker-compose, local /home and ~/workspace directories are mapped to persistent docker volumes respectivey. Thanks to this you can frequently update stellars-jupyterlab-ds container image without affecting your work
• Root Access: You have access to the local root account with sudo (password: password).
• TensorBoard Logs Directory: /tmp/tf_logs
• Ports:
  • TensorBoard: 6006
  • JupyterLab: 8888
• Conda Environments:
  • base - basic machine learning tools and packages
  • tensorflow - environment with all tools from base and latest stable tensorflow installed with NVIDIA GPU support
  • torch - environment with all tools from base and latest stable torch installed with NVIDIA GPU support. Additionally packages from ultralytics were installed for your YOLO projects

Tip: You can change envionment using conda activate env_name command. You can also set default environment either by setting CONDA_DEFAULT_ENV in docker-compose.yml file or in the ~/.profile file in your home directory

Tip: You don't need to run docker-compose build if you pull the image from Docker Hub. Running docker-compose up for the first time will automatically use the pre-built package if available.

Configuration Details

• ./build: Contains container build artifacts. Typically, you won't need to interact with this directory.
• ./.env: Contains the project name, used for naming volumes and the Docker Compose project.

Tip: Modify the /opt/workspace entry in the volumes: section of the docker-compose files to map to a different project location on your filesystem.

Platform Features

• JupyterLab 4+ (JupyterLab Homepage for reference)
• Extensions: Git integration, autocompletion, and other useful tools.
• Language Server Protocol (LSP): For Python autocompletion.
• Machine Learning Libraries: Keras, TensorFlow, Scikit-learn, SciPy, NumPy.
• Data Manipulation Libraries: Pandas, Polars.
• Visualization Libraries: Matplotlib, Seaborn.
• NVIDIA CUDA Support: GPU-accelerated libraries like CuPy, cuDF, and TensorFlow with GPU support.
• Miniconda: With enhanced terminal support in JupyterLab.
• Output Formats: HTML and PDF (WebPDF) generation.
• Memory Profiler: Monitor resource usage.
• Customizable File Mapping: Easily map your filesystem's project folder to the container.
• Persistent Settings: Configurable files and folders for JupyterLab settings, AWS credentials, and Git settings.
• Themes: IntelliJ dark theme with medium contrast.
• Favourites: Quick access for projects.
• Enhanced Terminal: Includes mc and other useful tools.
• TensorBoard: Pre-configured and running on port 6006.
• TensorFlow Visualization Extensions: For improved model insights.