Introduction

Running TensorFlow within a Docker container provides a secure and isolated environment to manage dependencies and configurations. It also streamlines collaboration by ensuring consistency across development, testing, and production environments. Pairing this setup with Jupyter Notebook offers an interactive way to write and debug machine learning code directly in the browser. This article will guide you through the process of setting up a TensorFlow-powered Jupyter Notebook using Docker.

Prerequisites

Before you begin, ensure you have the following installed on your system:

1. Docker: You can download it from the official Docker website.
2. Basic understanding of Docker: Familiarize yourself with Docker commands and container management.

Step-by-Step Guide

Step 1: Pull TensorFlow Docker Image

TensorFlow's official Docker Hub hosts pre-configured Docker images which include Jupyter Notebook. Start by pulling the appropriate image. For instance, to fetch the latest TensorFlow image with Jupyter enabled, use:

docker pull tensorflow/tensorflow:latest-jupyter

Step 2: Run Docker Container with Jupyter

Once the image is downloaded, run the container using the following command which maps Jupyter’s default port and mounts a local directory for data persistence:

docker run -it --rm -p 8888:8888 -v $(pwd):/tf tensorflow/tensorflow:latest-jupyter

Explanation:

• -it: Starts the container in interactive mode.
• --rm: Automatically removes the container when it stops running.
• -p 8888:8888: Maps the local port 8888 to the container’s port 8888, the default port for Jupyter Notebook.
• -v $(pwd):/tf: Mounts the current directory to /tf within the container to ensure data saved in Jupyter notebooks is retained on your local machine.

Step 3: Access Jupyter Notebook

Upon running the container, the output in the terminal will include a URL with a token for Jupyter. It looks something like:

http://127.0.0.1:8888/?token=abcd1234efgh5678

Copy and paste this URL into your preferred web browser to access Jupyter Notebook.

Step 4: Run TensorFlow Code in Jupyter

With Jupyter Notebook up and running, you can create a new notebook and begin executing TensorFlow code. Here is an example snippet to verify TensorFlow's installation:

import tensorflow as tf

print("TensorFlow version:", tf.__version__)

Key Configuration Points

Additional Configurations

1. Running with GPU

If you wish to leverage GPU capabilities, ensure you have the appropriate NVIDIA drivers and nvidia-docker installed. Run the container with:

docker run --gpus all -it --rm -p 8888:8888 -v $(pwd):/tf tensorflow/tensorflow:latest-gpu-jupyter

2. Customizing Jupyter Server

You can pass additional environment variables or configuration scripts to tailor the Jupyter server further. For example, to disable token authentication (not recommended for production), you could modify the Docker command:

docker run -it --rm -p 8888:8888 -e JUPYTER_TOKEN='' -v $(pwd):/tf tensorflow/tensorflow:latest-jupyter

Conclusion

Setting up TensorFlow in a Docker environment with Jupyter Notebook adds a robust layer of flexibility and portability to your machine learning workflows. This setup is well-suited for both experimental development and deploying stringent production environments. By following these steps, you're equipped to harness the power of TensorFlow and Jupyter seamlessly and efficiently.