Neural Network Docker Example

Example of how to use docker to run a neural network on NVIDIA GPUs

Structure of the project

The project is structured as follows:

• Dockerfile: Docker creation file containing parameters and definitions to run the algorithm
• train.py: Contains the script to train the model.
• LICENSE: code use license

Instructions for running the code

1. Build the image

Once a Dockerfile has been defined for the project to be executed, the next step is to create the container image. The following command is employed to generate the image:

docker build -t [image-name] .

It is necessary to select an image name ([image-name]) and to place it in the Dockerfile folder when the build is initiated.

2. Create and run the container

Once the image has been created, the container can be run. This is achieved by executing the following command:

docker run --gpus all -v my-host-path:/usr/src/app/logs [image-name] > my-host-path/log.txt 2>&1

This command instructs Docker to execute a new container on the available GPUs. A volume is mounted, connecting a folder on the host with a specific folder in the container. Standard output (stdout) and standard errors (stderr) are redirected to a log file.

Once the process has been completed, the container can be deleted and the results can be viewed in the host folder that was specified.