Environment

Hardware

Google Cloud Platform

• Primarily used for training models.
• Free $300 trial credits.
• Many different types of services available.

Compute Engine

• Great for creating scalable servers with high CPU count.
• Maximum of 24 CPU cores.
• Direct SSH access through pop up.

Firewall Configuration

• Default configuration blocks all incoming connections.
• Certain ports need to be opened for different web servers.
• Suggested configuration is shown above.

Software

Ubuntu

• Preferred over Windows due to strong data science support. (eg. TensorFlow only supports Windows on Python 3 and not Python 2.7)
• Open-source licensing.

Jupyter

• Popular among data scientist.
• Great platform for the team to share code and monitor models.
• Allows the team to work on a remote server. (More powerful than local PC)
• Notebooks allow for interactive output and rich formatting
• Open-source licensing.

Installation & Configuration

1. Install Jupyter

   sudo pip install jupyter

2. Generate configuration file

   sudo .local/bin/jupyter_notebook --generate-config

3. Configure Jupyter

   sudo vi .jupyter/jupyter_notebook_config.py

Edit the following fields in the configuration file:

c.Notebook.allow_password_change = True
c.NotebookApp.allow_root = True
c.NotebookApp.ip = "*"

4. Change password of Jupyter notebook

   sudo .local/bin/jupyter_notebook password

5. Run Jupyter notebook

   tmux

   sudo .local/bin/jupyter_notebook