Getting Started with Fast.ai MOOC in AWS

Fast.ai provides a deep learning course series. Program is taught by Jeremy Howard. There is a getting started page to follow instructions. However installation did not happen easy for me. I want to share some problems I have encountered and solutions of them.

Firstly, I opened this instructional video. My plan was trying to imitate instructor. Type p2 instance is not available instantly, you need to request it from AWS. Return of AWS takes time, so you can use type t2 instance during this interval. Let’s come back our computer, we need to install Anaconda with Python 2.7 and Cygwin . Instructional video will warn you related with wget. There is the first hint of mine. You should also be careful about ssh. You should select openssh and not skip to install it. If you skip, you get this message at the last step to connect your instance:

-bash: ssh: command not found

If you skip it nevertheless, you should find setup file which you installed a few minutes ago. Duplicate the previous clicks, and select openssh packages and install. No need to close terminal during these operations.

Ok, let’s turn back video. After the installation of Cygwin, we will create a user in AWS Console. Key points are giving the administrator privileges and saving the credentials of access key id and secret key.They will need to connect our instances.

Hereafter, we jump to Cygwin and write:

pip install awscli

Everything is seen good but don’t be fooled. The next step could be this:

C:\Users\serdar\Anaconda2\python.exe: can’t open file
[Errno 2] No such file or directory

There is a solution page for this problem. However it didn’t help me. My solution came from here. Updating bash_profile with PATH of python location made me happy.

$ echo “PATH=\$PATH:/cygdrive/c/Python27” >> .bash_profile
$ source .bash_profile

Ok. That’s it. AWS configuration is available for now. Enter the credentials you saved before, selected region and output format.Now we need to start ready-to-use instance by script. It is linked to here. Key point is using the raw code link. Example for t2:

https://raw.githubusercontent.com/fastai/courses/master/setup/setup_t2.sh

By using wget, we get it for instance and typing:

bash setup_t2.sh

will setup our instance.

When setup is completed, it will print connection information for instance. Note that this info. My ssh problem is seen this point. You will use “ssh -l” and connection address. Finally, you are in! You can open Jupyter, and connect that by using “DNS of instance:8888” by browser of your computer. Don’t forget to stop your instance after finishing your job for the sake of your budget. Have a nice MOOC..