setting_up_environment_for_fastai.md

Setting up an Enviornment for Fastai

TOC

• Google Colab Setup
• Local Setup

Useful links

• The fastai forums
• Notebook on how to use Google Colab Setup for fastai

Google Colab Setup

This is by far the easiest route. For each lesson, just open the notebook at select File > Save copy in Drive ...

Note, you may have to use Google Chrome. Google Colab did not work in Safari for me.

• Lesson 1: lesson1.ipynb
• Lesson 2: lesson1.ipynb, lesson1_breeds.ipynb
• Lesson 3: tmp_lesson1-quick.ipynb, keras_lesson1.ipynb, conv-example.xlsx, entropy_example.xlsx, lesson2-image_models.ipynb, lesson3-rossman.ipynb
• Lesson 4:
• Lesson 5:
• Lesson 6:
• Lesson 7:

Then run the "setup" cells that I have added to the beginning of each notebook. Thats it! To use a GPU, go to Edit > Notebook settings > Hardware accelerator > GPU.

Local Setup

Installing python / conda

Arguably the easist way to manage your python / deep learning stack is to use conda, specifically, miniconda. Installation instructions can be found here, but essentially:

# download the installer
curl https://repo.continuum.io/miniconda/Miniconda3-latest-Linux-x86_64.sh --output ~/miniconda.sh

# run it
bash ~/miniconda.sh

Setting up a fastai enviornment

Follow the instructions here to setup an enviornment for following along with the fastai course (either GPU or CPU). Essentially:

# make sure conda and pip are up-to-date
conda update conda

# create a virtual enviornment with everything we need for running the lessons 
git clone https://github.com/fastai/fastai.git
cd fastai
conda env create -f environment.yml

# activate that enviornment
conda activate fastai

There are some additional steps not present in the instructions above. You need to install nb_conda, which allows you to use your virtual environment from within a notebook:

conda install nb_conda

and you need to downgrade one of the pytorch packages (this is likely temporary!)

pip install torchtext==0.2.3

Using the notebook

To use the notebook, you have to activate your enviornment:

source activate fastai

move into the fastai repo

cd path/to/fastai

and launch jupyter notebook

jupyter notebook

Make sure that you select the right kernel! (Kernel > Change kernel > Python [conda:fastai])

Accessing your jupyter notebook remotely

Following the instructions here for a secure setup.

For a less secure setup,

1. On the host machine:

source activate fastai
cd path/to/fastai
jupyter notebook

2. On the local machine:

8888:localhost:8888 <username>@<host>

the notebook should be accesible in your browser at localhost:8888.