Add Introduction to PyTorch tutorial #2440
Conversation
github-actions
bot
added
the
documentation
| { | ||
| "cells": [ | ||
| { | ||
| "cell_type": "code", |
There was a problem hiding this comment.
Would like to start formatting the copyright in code cells, as tools like ruff can only check code cells: astral-sh/ruff#12802
| "source": [ | ||
| "# Introduction to PyTorch\n", | ||
| "\n", | ||
| "_Written by: Adam J. Stewart_\n", |
There was a problem hiding this comment.
Would like to start giving people credit for the tutorials they write, especially for the case study tutorials. This is intended to be the first person who wrote the tutorial, we wouldn't append other names except in the case of major revisions.
| "id": "eb5dc7e8-6cb3-4457-83ad-7fa5aef8ea0c", | ||
| "metadata": { | ||
| "nbmake": { | ||
| "mock": { |
There was a problem hiding this comment.
The tutorial runs for 100 epochs, but we mock it to only run for a single epoch during testing.
There was a problem hiding this comment.
This new tutorial only takes 9.69s to test.
| "metadata": {}, | ||
| "outputs": [], | ||
| "source": [ | ||
| "evaluate(test_dataloader)" |
There was a problem hiding this comment.
Final accuracies are 65% on validation and 85% on testing. Pretty good for only 100 images. We are nowhere near plateauing, and could run for longer to get higher accuracies, but I think this proves the point.
|
Would not put "self driving cars" as an example of DL |
|
Why not? I think it's a good example of a vision problem that would be familiar to a general audience. |
The first of many tutorials introduced for #2418
This tutorial is designed for remote sensing folks with little to no background in deep learning or PyTorch. It provides a foundation upon which all later tutorials will build.