Simply put, GCE is a service that lets you create and run VMs on Google infrastructure. The GCE allows you to launch spin up VMs with the right operating system, size, RAM, and the appropriate number of CPUs or GPUs for your needs. It is the equivalent of AWS EC2.

The GCE was announced on June 28, 2012, at Google I/O 2012 and made available to the general public on May 15, 2013. Compared to AWS EC2, an equivalent product, the GCE is a rather new service:

The following extracts from the release notes timeline illustrate the rapid evolution of the GCE service from a simple contender to a fully fledged player in the Cloud computing domain:

• May 15, 2013: GCE is available for everyone.
• August 6, 2013: GCE launches load balancing.
• December 3, 2013: GCE is announced as being production ready. Users can now feel confident using Compute Engine to support mission...

So, now we have a VM running and we're able to send files to it, connect to it, and modify it. Everything is ready for us to set it up for data science!

We will install the Python Miniconda stack from continuum, much smaller than the full Conda distribution. Do SSH into your instance.

1. Install the mini sudo apt-get update sudo apt-get install bzip2 wget from https://repo.continuum.io/miniconda/Miniconda2-latest-Linux-x86_64.sh using the following command:

bash Miniconda2-latest-Linux-x86_64.sh

2. And then install the Python stack with conda:

$ conda install scikit-learn pandas jupyter ipython

Don't forget to do this:

$ source .bashrc

To launch a Jupyter Notebook in your instance and access it over the Web, you need to promote the ephemeral external IP address provided by default to your VM to a static external IP.

You also need to make sure that your instance is accepting HTTP and HTTPS traffic. For that, go to your VM page, edit it, and check the following checkboxes:

Since your Jupyter Notebook is open to all traffic on the web, you password-protect it:

1. Generate a configuration:

$ jupyter notebook --generate-config

2. And add a password with:

$ jupyter notebook password

More info on how to protect your public Notebook, including using ssh keys and adding encryption, is available at http://jupyter-notebook.readthedocs.io/en/latest/public_server.html.

3. Launch your Jupyter Notebook with:

$ jupyter notebook --ip=0.0.0.0 --port=8888 --no-browser &

This generates a token and the...

The following are a few interesting articles on setting TensorFlow on GCE:

• Running distributed TensorFlow on Compute Engine: https://cloud.google.com/solutions/running-distributed-tensorflow-on-compute-engine
• Jupyter + TensorFlow + Nvidia GPU + Docker + GCE: https://medium.com/google-cloud/jupyter-tensorflow-nvidia-gpu-docker-google-compute-engine-4a146f085f17
• Using a GPU and TensorFlow on GCP: https://medium.com/google-cloud/using-a-gpu-tensorflow-on-google-cloud-platform-1a2458f42b0
• Running Jupyter Notebooks on GPU on Google Cloud: https://medium.com/google-cloud/running-jupyter-notebooks-on-gpu-on-google-cloud-d44f57d22dbd

Some docker-related resources:

• Docker config to create machines on GCE: https://docs.docker.com/machine/drivers/gce/
• Logging with stackcriver: https://medium.com/google-cloud/how-to-log-your-application-on-google-compute-engine...

The GCE is a GCP core service, offering a wide variety of scalable VMs based on different OSes. The multiplicity of available OSes, the range of machines with CPUs, GPUs, small to huge disk space, and RAM make the GCE a powerful Cloud environment adapted to a wide variety of projects and contexts.

In this chapter you learned:

• How to create, launch, back up, modify, and access multiple VMs
• The different parameters and variables related to a VM
• How to access and use the Google Shell
• How to use the gcloud CLI to carry out the same operations in the GCP
• How to install a data science Python stack
• How to launch a Jupyter Notebook

There are many more possibilities offered by the power and flexibility of the GCE that we haven't covered. Hopefully, by the end of this chapter, you should feel comfortable working with instances that are appropriate for your projects in a data...