Best Practice: From Bare Metal to Docker Images¶
Some Researchers do data science on bare metal. The term bare-metal relates to connecting to a server and working directly on its operating system and disks.
This is the fastest way to start working, but it introduces problems when the data science organization scales:
- More Researchers mean that the machine resources need to be efficiently shared
- Researchers need to collaborate and share data, code, and results
To overcome that, people working on bare-metal typically write scripts to gather data, code as well as code dependencies. This soon becomes an overwhelming task.
Why Use Docker Images?¶
Docker images and containerization in general provide a level of abstraction which, by large, frees developers and Researchers from the mundane tasks of setting up an environment. The image is an operating system by itself and thus the 'environment' is by large, a part of the image.
When a docker image is instantiated, it creates a container. A container is the running manifestation of a docker image.
Moving a Data Science Environment to Docker¶
A data science environment typically includes:
Training data is usually significantly large (from several Gigabytes to Petabytes) and is read-only in nature. Thus, training data is typically left outside of the docker image. Instead, the data is mounted onto the image when it is instantiated. Mounting a volume allows the code within the container to access the data as though it was within a directory on the local file system.
The best practice is to store the training data on a shared file system. This allows the data to be accessed uniformly on whichever machine the Researcher is currently using, allowing the Researcher to easily migrate between machines.
Organizations without a shared file system typically write scripts to copy data from machine to machine.
Machine Learning Code and Inputs¶
As a rule, code needs to be saved and versioned in a code repository.
There are two alternative practices:
- The code resides in the image and is being periodically pulled from the repository. This practice requires building a new container image each time a change is introduced to the code.
- When a shared file system exists, the code can reside outside the image on a shared disk and mounted via a volume onto the container.
Both practices are valid.
Inputs to machine learning models and artifacts of training sessions, like model checkpoints, are also better stored in and loaded from a shared file system.
Any code has code dependencies. These libraries must be installed for the code to run. As the code is changing, so do the dependencies.
ML Code is typically python and python dependencies are typically declared together in a single
requirements.txt file which is saved together with the code.
The best practice is to have your docker startup script (see below) run this file using
pip install -r requirements.txt. This allows the flexibility of adding and removing code dependencies dynamically.
ML Lifecycle: Build and Train¶
Deep learning workloads can be divided into two generic types:
Getting your docker ready is also a matter of which type of workload you are currently running.
With "build" you are actually coding and debugging small experiments. You are interactive. In that mode, you can typically take a well known standard image (e.g. https://ngc.nvidia.com/
Start a docker container by running:
docker run -it .... "the well known image" -v /where/my/code/resides bash
You get a shell prompt to a container with a mounted volume of where your code is. You can then install your prerequisites and run your code via ssh.
You can also access the container remotely from tools such as PyCharm, Jupyter Notebook, and more. In this case, the docker image needs to be customized to install the "server software" (e.g. a Jupyter Notebook service).
For training workloads, you can use a well-known image (e.g. the TensorFlow image from the link above) but more often than not, you want to create your own docker image. The best practice is to use the well-known image (e.g. TensorFlow from above) as a base image and add your own customizations on top of it. To achieve that, you create a Dockerfile. A Dockerfile is a declarative way to build a docker image and is built in layers. e.g.:
- Base image is nvidia-tensorflow
- Install popular software
- (Optional) Run a script
The script can be part of the image or can be provided as part of the command line to run the docker. It will typically include additional dependencies to install as well as a reference to the ML code to be run.
The best practice for running training workloads is to test the container image in a "build" session and then send it for execution as a training Job. For further information on how to set up and parameterize a training workload via docker or Run:AI see Converting your Workload to use Unattended Training Execution.