In this repo, we aim to replicate the notebooks available in fast.ai repository but in such a way that they are compatible with Google's Colab notebook environment that has a free to use GPU backend (which you will need to enable via Runtime -> Change runtime type). The notebooks are usually self contained, i.e. it will install all the required packages and download the relevant datasets using inline shell scripts since the kernels get destroyed after 12 hours of use (I am hoping that they will eventually allow storage and retrieval of data from Google Drive, which is only an option for notebooks at the moment). As a result, the instructions below are not relevant for Colab.

NOTE: Only go ahead with this route if you have a decent nvidia GPU (compute capability 3+) as anything lower does not support CUDNN and you don't want to do deep learning on a CPU (your iteration cycle will suffer).

The fast.ai python package requires Python 3.6 or higher.

On Ubuntu:

sudo add-apt-repository ppa:jonathonf/python-3.6
sudo apt update
sudo apt install python3.6

On a Mac:

Install the latest HomeBrew by running the following command:

/usr/bin/ruby -e "$(curl -fsSL
https://raw.githubusercontent.com/Homebrew/install/master/install)"

Then,

brew install python3

If you have Anaconda installed, you may need to removing it from your $PATH bycommenting it out from your ~/.profile or ~/.bash_profile.

git clone [email protected]:strange-labs-uk/fast.ai
cd fast.ai
ENV=~/Envs/ML-Python3
mkdir -p $ENV
virtualenv -p python3 $ENV
echo "#! /bin/bash" > env.sh
echo ". $ENV/bin/activate" >> env.sh
. env.sh
pip install -r requirements.txt
jupyter notebook

Run the following script to download data which will automatically download and unpack images of dogs and cats to ~/Datasets/dogscats.

./get_dataset.sh