zupo / fast_ai

Notes on doing Practical Deep Learning For Coders, Part 1

Notes

• no need to create an API user, use the credentials I sent in the email
• after you use setup_p2.sh script to create an instance, go to AWS Console, select it, click Action, click Add/Edit Tags, prepend your nickname to Name of instance so that you can easily find it on the list of instances when others add their own
• reboot after first SSHing into instance, so nvidia is propertly set up

Don’t break your dev python env by not installing Anaconda as root Python

• install anaconda with GUI instaleld
• comment out anaconda PATH in ~/.bash_profile
• create a conda isolated env: ~/anaconda/bin/conda create --name course
• activate the env: source ~/anaconda/envs/course/bin/activate

Let's verify. Run the following in the terminal:

(root) course$ which python
/Users/zupo/anaconda/bin/python

Run this in a new terminal window. Python path should be to your system Python, not to Anaconda's:

course$ which python
/usr/local/bin/python

Don't overwrite existing AWS config

If you already used awscli in the past, you have configuration and credentials already set, and blindly following the course will make you overwrite them.

• edit ~/.aws/config and change [default] to [profile original] (or similar)
• do the same for ~/.aws/credentials
• now whenever you need your original config/credentials, run aws with --profile original

Running locally

$ conda install bcolz
$ conda install theano
$ pip install keras==1.2.2
$ pip install np_utils

$ jupyter notebook