Working environments

On your machine

The most straightforward way to develop: installing every library on your machine. This method is still relevant today, and has its perks, but requires time and a bit of knowledge on computers and stuff... This is why we don't recommend it for beginners, who just may be looking for a quick way to test some code.

• Advantages:
  • Run locally, no internet connection needed
  • Code fast once installation is done
  • Easily connects building blocks
  • Heat up the room in winter
• Disadvantages:
  • Long and complicated installation
  • Expensive computational power
  • Hard to share your code

Colab

This new tool developed by Google is a collaborative Jupiter notebook. Coming along with all major libraries pre-installed, a reasonable computing power, and close to no time to set up, it is by far the best environment for experimenting with deep learning. On the long run, the limitations of a free service encourages the use of more robust tools, like the following on this list.

• Advantages:
  • Easy to set up
  • Easy to share
  • Powerful free computational power
  • Good set of tools (Jupyter notebook, git, ...)
• Disadvantages:
  • Hard to load data
  • Reset every day
  • Not 100% secure

Paperspace

Please review me =)

• Advantages
  • Easy to set up
  • Good computational power with low prices
  • Good set of tools (Jupyter notebook, git, ...)
• Disadvantages:
  • Expensive on the long run
  • Not 100% secure

AWS and GCE

If you are looking for the tools used by Apple and Netflix, you found the right place. These will follow you from experimenting, coding to deploying and even marketing your solution. They also provide you with powerful GPU that you can rent for a few hours at a reasonable price (~1$/hour). Learning how to use these tools is long and complicated, but the knowledge you spend hours acquiring may very well serve you all your career. We encourage using these tools for long projects that require either big computational power or to be 100% secure.

• Advantages:
  • Easy to set up
  • Easy to deploy API
  • Huge computational power with low prices
  • Good set of tools (Jupyter notebook, git, ...)
• Disadvantages:
  • Inscription and first launching are complicated
  • Expensive on the long run
  • Steep learning curve