Fastest way to setup Fast.ai v3 2019 course notebooks — using Google Colab and Clouderizer

Prakash Gupta
Mar 16, 2018 · 3 min read

Inspired by this post from Manikanta Yadunanda, our team at Clouderizer successfully ported our fast.ai project template to work seamlessly on Google Colab.

Recent changes in Google Colab has made running Fast.ai notebooks easier but still there are few gotchas.

  1. Additional code cells, for dependency setup, need to be injected in each notebook and run every time we return to work.
  2. By default our code changes, models and dataset are lost every time Colab instance gets reset. In order to persist them to Google Drive, we need to
    a. Explicitly put Google Drive mount code in every notebook.
    b. Go through Google OAuth token workflow to grant notebook access to our Google Drive.
    c. Change code in notebook everywhere to use mounted google drive path, instead of default paths.
  3. Colab notebooks, while being free and awesome in lots of ways, still lack few features in comparison with Jupyter Notebooks.

Clouderizer attempts to address all the above hassles and make running Fast.ai v3 notebooks easy. Below are the steps to setup fast.ai course notebooks on Google Colab.

Following pre-requisite, one time, steps are needed.

  1. Sign up and request access to Google Colab.
  2. Sign up for Clouderizer.
  3. Login to Clouderizer console. On first login, you will be prompted to link your Google Drive with Clouderizer. Follow on-screen instructions to do so.
  4. Now from Clouderizer console, go to Community Projects. Search for fast.ai template and clone it.
  5. On project wizard, select default options on all screens and save.

Now every time you need to start fast.ai notebooks on google colab GPU, follow below 2 steps

  1. From Clouderizer console, press Start on fast.ai project created earlier. This will ask you to select the platform where you wish to run this project. Select “Google Colab” here. This should create a Colab notebook for your project. Press “Launch Notebook Now”. This will open the Colab notebook in a new tab.
Select Colab to run your project on Google Colab
  1. Run the code block in Colab notebook.
Run code block on Colab to kick start Clouderizer project

Thats it! This will trigger an automated fast.ai course environment setup, latest code download and dogscats dataset download. You can go back to Clouderizer console and track the progress of this setup. Once setup is complete, project status becomes Running and Jupyter Notebook button becomes available. Clicking on Jupyter button, will open Jupyter notebook with fast.ai github code.

Project Started

IMPORTANT NOTE

  1. We are giving away one month worth coupon code for Fast.ai students. Please use coupon code FASTAI2019
  2. In case while working on JupyterLab or remote terminal, you get 502 Gateway Error, this is due to Colab instance getting terminated behind the scene. Please note that Google Colab instances are volatile. They get terminated every 12 hours, or can get terminated much sooner in case of high demand and idle sessions.
    While your Clouderizer project is running on Colab notebook, make sure you don’t close Colab notebook tab. Preferably, keep it in foreground on one of your secondary displays. This helps in preventing pre-mature shutdown of Colab instance.

Prakash Gupta

Written by

Passionate about designing and developing scalable software platforms.