Intro to Deep Learning Tutorial Project

These materials are for the 'Introduction to Deep Learning' tutorial. The goal of this tutorial is to get hands-on with PyTorch and GPU Deep Learning (DL). It is specifically targeted toward attendees who may be familiar with the concepts of DL, but want practical experience. Familiarity with Python and typical ML packages (e.g. pandas, numpy, sklearn) is expected.

At the end of this session, you will understand how to:

• Build some common DL architectures in PyTorch
• Evaluate and improve the performance of DL models
• Take advantage of more compute (and when you should do so)

Project Contents

• admin-setup folder - contains code for downloading data and other setup for the admin copy of this project
• 1-Load-Data-PyTorch.ipynb - Tutorial notebook for loading and visualizing data in PyTorch
• 2-Build-Model.ipynb - Tutorial notebook for training a model with transfer learning
• 3-Improve-Performance.ipynb - Tutorial notebook for improving model training speed
• 4-Hyperparameters-Tuning.ipynb - Tutorial motebook for improving model accuracy

Project setup

The below recommended flow for setting up this project assumes it has been uploaded to a Domino deployment as a public project, and a Domino admin has set up the prerequisites listed below.

1. Fork the project
2. In your forked copy of the project, mount the data as follows
   • go to the Materials section in the left navigation of the project and click Data
   • go to the Domino Datasets tab
   • click Mount Shared Dataset
   • choose Serengeti4kImages from the searchable dropdown and click Mount Dataset
3. Launch a Workspace to explore the notebooks.
   • go to the Run section in the left navigation of the project and click Workspaces
   • click Create New Workspace
   • choose Deep Learning Tutorial from the workspace environment drop-down
   • choose JupyterLab as the IDE
   • choose a GPU tier from the hardware tier drop-down (the smallest available GPU is fine)
   • click Launch Now

Prerequisites

These should be completed by a Domino admin.

Dataset setup

The public copy of this project should have a Dataset called Serengeti4kImages available to mount to other projects. See admin-setup/Download-Workshop-Data.ipynb for details on downloading the data into the Dataset.

Compute Environment

There should be a Compute Environment called Deep Learning Tutorial available with the following properties.

• Custom base image: quay.io/domino/ngc-pytorch:20.12-py3
• Dockerfile instructions: RUN pip install line_profiler
• Pluggable workspace tools:

jupyterlab:
  title: "JupyterLab"
  iconUrl: "/assets/images/workspace-logos/jupyterlab.svg"
  start: [  /opt/domino/workspaces/jupyterlab/start ]
  httpProxy:
    internalPath: "/{{ownerUsername}}/{{projectName}}/{{sessionPathComponent}}/{{runId}}/{{#if pathToOpen}}tree/{{pathToOpen}}{{/if}}"
    port: 8888
    rewrite: false
    requireSubdomain: false
vscode:
  title: "vscode"
  iconUrl: "/assets/images/workspace-logos/vscode.svg"
  start: [ "/opt/domino/workspaces/vscode/start" ]
  httpProxy:
    port: 8888
    requireSubdomain: false

Equivalent environments will work as well. (For example, NGC pytorch images used with the option to Automatically make compatible with Domino, or any other environment with working pytorch installed.) If the line_profiler package is not installed, it can also be installed within the tutorial notebook where needed.

Reference material

• Data location - https://www.zooniverse.org/projects/zooniverse/snapshot-serengeti
• Article - https://www.pnas.org/content/115/25/E5716
• Article code repo - https://github.com/Evolving-AI-Lab/deep_learning_for_camera_trap_images