fastai

These are a set of scripts to help with the Fast.ai DL course

Instructions

The main instructions for setting up AWS for the course are here. The course have you setup a GPU base AWS instance that rates at $0.90/hr. The scripts in this project allows setting up an additional t2-micro instance that helps when doing non-DL workloads, like mungings.

Requirements

The rest of these instructions assumes that you have AWS credentials and installed the AWS cli. See video.

Initial setup

This project assumes that you have a copy of the Fast.ai courses project. To get that code, run make init.

Create necessary images

The current scripts only generate the GPU based images.

Run make setup to execute the Fast.ai provided scripts to setup the AWS instance. Verify in AWS console that the instance was successfully created.

Creating the t2-micro instance is more of a manual process. This can still be easily done by running the Fast.ai provided scripts. The goal is 2 instances that can share the same Volume, on free, the other expensive. Follow these steps:

1. Edit courses/setup/setup_t2.sh to create a t2-micro instance rather than a t2.xlarge.
2. Run courses/setup/setup_t2.sh
3. Verify that the instance was created in AWS console
4. Change the name of the new instance to be different than the gpu one. (i.e call it fast-ai-munging-machine)
5. Stop the instance if it is running
6. Go to the Volumes section in AWS Console and Detach and Delete the volume that was created for the t2-micro image. HINT: Most likely the one with the most recent date.
7. Detach the volume assigned to the p2-xlarge instance and Attach it to the t2-micro. Make sure that you set the Device to /dev/sda1. Note that for all this to work, both instances need to be stopped. (TODO: Automate all this!)

NOTE: Moving the volume to the correct instance is manual.

Start/Stoping an instance

Use make start and make stop to control the state of the instance. The first time any of this targets are run, make will fail if it does not know what instance is currently used. This is a one time setup by setting the variable INAME to the Name or part of the Name of the instance. If using a partial name, make sure it uniquely identifies the instance. For example:

make start INAME=munging

This is cached and not needed for other commands. To clear things, run make clean.

Connection to the instance

Run make connect to ssh into the current instance.

Opening the Jupyter notebook

Run make nb and the Chrome will open the right URL. Note that this command assumes OSX.

Resources

• AWS API - http://docs.aws.amazon.com/cli/latest/reference/ec2/describe-instances.html