Web Analytics Made
Easy - StatCounter

Notes for running the FastAI 2020 Jupyter notebooks.

To get started, I tried running the code for the first 9 chapters of the Jupyter notebooks for the FastAI course. There’s some little tricks that I had to look up, so I’ll just document them all here.

Roadmap

Getting started, browser.

First, concerning the browser to run the Jupyter notebooks in, I hit some snags running the notebooks in Firefox in Ubuntu. I then used Chromium and then had no issues. But now I cannot reproduce this issue, so take it with a grain of salt.

Chapter 2, Microsoft key in Azure.

In Chapter 2, you’re instructed to register at Microsoft and get a key. Ehhh? Here, I’ll mention that the forums are your best chance of resolving issues quickly. Here’s my post on how to register at Microsoft Azure and get the Bing Search (free version) set up. I did need a credit card to register.

Chapter 7.

Using an older version of the FastAI Docker images, I ran into a prior bug. If you happen to run into it too, here’s the forum fix. BUT it was fixed with this bug report, so fresh versions of FastAI shouldn’t have this issue.

Chapter 9, Kaggle.

For Chapter 9, you’ll have to sign up at Kaggle, and also need a key to run the code. If you started by running the notebook first, it will have created the subdirectory .kaggle under your user’s workspace, with an empty kaggle.json inside. (Then you’ll get an error).

At Kaggle, go to Your Account->API->Create API Token. You’ll be presented with a .kaggle/kaggle.json -file – save it in your home directory, or copy-paste the contents into your empty file at the same location. Just be sure the permissions are correct for your home directory.

While you’re at Kaggle, you also need to accept the rules of the competition to download the dataset, so search for ‘Bluebook for Bulldozers’ or here, go to ‘Rules’ and accept.

One more thing! There’s a typo. The forum has the current fix for the m_rmse line.

Two code blocks down, there’s another problem at the to_nn = TabularPandas... line. The forum fix is here.

OKAY. Now, try running it.

To be continued.

I do plan to continue running the course notebooks further than chapter 9, but that time is not now. So check back later for updates to this page and also the Dockerfile!

© Amy Tabb 2018-2020. All rights reserved. The contents of this site reflect my personal perspectives and not those of any other entity.