PyTorch + FastAI: Deep Learning setup on Paperspace

1. Sign up on paperspace.com and verify your email to activate your account.
2. Create a new P5000/P6000 machine with at least 100 GB storage and a public IP. Chose the Ubuntu 16.04 template.

https://www.paperspace.com/console/machine/create/

3. Start the machine and open a terminal in the browser.

4. Run the following command to setup up the tools used for development viz. CUDA, Anaconda, PyTorch and FastAI:

curl http://files.fast.ai/setup/paperspace | bash

This can take a long time. When it’s done, make sure to reboot the machine.

5. Enter the fastai directory and start Jupyter notebook:

cd ~/fastai
jupyter notebook

6. Copy the URL shown on the screen, and change localhost to the public IP of your machine, then paste it a new browser tab. This should open up the Jupyter notebook:

7. Go to courses/dl1/ and create a new notebook called catdog.ipynb.

8. Copy and execute the following code:

If you see something similar to the following output, you’re good to go:

At this point you’ve successfully trained a cats vs. dog classifier with over 99% accuracy. 🎉🎊🏆

9. If you’re new to deep learning, or are totally lost about what just happened, you can now start watching the first video from Practical Deep Learning for Coders, starting around the 13-minute mark:

This article basically summarizes the first 13 minutes of the above video as text for easier reference.