MyDocker: How to build and use a Docker image for Deep Learning development.

Problemática

1. If 2 people are working on the same project but with 2 different environments, there may be dependency version issues.
2. Some packages are optimize for a special OS.

I. What is Docker

Docker is a manager of container.

Container:

• Isolates app from each other

• Shares the same kernel – kernel shares resources with the host and interacts with the containers

Advantages

• Quicker to launch (no os to boot)

• Portability (less dependencies between processed layers)

• Efficiency

II. Run a container

1. Pull an image from Docker hub: docker login then docker pull

2. Check the image: docker images

3. Create a container: docker run –it nom Few options:

   INSTRUCTION	SIGNIFICATION
   -i	Interactive mode
   -p	Publish a container᾿s port or a range of ports to the host format (-p 8888:4000)
   -t	Allocate a pseudo-tty
   -v path/to/volume	Bind mount a volume

4. Stop the container: docker stop id_container nom

5. Interact with a container with shell: docker run –it nom_container /bin/bash

6. Save the container after modification: docker commit id_container nom

7. Stop the container

8. Confirm the container has been stop: docker ps –a

9. Delete the container: docker rm id_container

III. Build your own container

1. Create a directory
2. Create your Dockerfile

Commands:

INSTRUCTION	SIGNIFICATION
FROM	Set base image
LABEL	Add metadata
COPY	Copy files/directories into the image
ENV	Set environment variable
EXPOSE	Informs Docker that the container listens on the specified network ports at runtime
WORKDIR	Set working directory
RUN	Execute shell commands in a new layer
ENTRYPOINT	Configure container to run as executable
CMD	Provide default for executing container

Example

3. Create hello_world.py: print("Hello world!")
4. Build your image: docker build –t nom .
5. Create the container: docker run nom
6. Upload to docker hub:
   • docker login
   • docker tag image username/repository:tag (example: docker tag friendlyhello john/first_docker:test)
   • docker push username/repository:tag

IV. Exercise

1. Create a directory

2. Create your Dockerfile in this new directory.

   FROM ubuntu:latest

   RUN apt-get update && apt-get install -y python3 python3-pip

   COPY requirements.txt .

   RUN pip3 install jupyter

   RUN pip3 install -r requirements.txt

   RUN useradd -ms /bin/bash jupyter

   USER jupyter

   WORKDIR /home/jupyter

   COPY hello_world.txt .

   ENTRYPOINT ["jupyter", "notebook", "--ip=*"]

3. Create requirements.txt with Tensorflow, Keras.

4. Create hello_world.txt

5. Build your image: docker build -t jupyter .

6. Create a new container: docker run -it -p 8888:8888 jupyter /home

7. Consulte the results on your browser (you might need to change your ip address)