Tensorflow In Practice Specialization

This repository contains the work and other tools for working through Cousera's Tensorflow in Practice Specialization.

There is multiple ways to work with this repo:

1. Using Google Colabs tool, an awesome free tool provided by Google to do Research & Development with notebooks.

2. Docker containers :) as I love using docker containers for almost everything I work with. I've created a set of tools you could use for your own projects apart from this one.

Docker Container Setup

If you have:

• GPU
• Ubuntu / Debian Linux Distribution

./scripts/install-docker-nvidia.sh

You should see an output like this if everything went well :) A table

TODO: Add picture of output here

Install docker container

docker build -f scripts/Dockerfile -t tf_practice_gpu .

If you have:

• CPU
• Ubuntu / Debian Linux Distribution

docker build -f scripts/Dockerfile-cpu -t tf_practice_cpu .

Docker build arguments:

• -f path_to_dockerfile, pass the path to your dockerfile
• -t tag, will set a tag to your image, in our case, tf_practice_gpu to be able to run the container that is only for gpu.

Starting your docker container

When I started building docker containers for everything, the idea was to spin it up those with the data I wanted :)

In this case, you will be able to run your container mounting any folder in the volume named as: /data

For example, Imagine you are going through lesson 1 of the specialization and you have a folder named: l1_prog_paradigm, running your container should be as easy as executing the following steps:

# enter the folder 
cd l1_prog_paradigm

# Docker container run enabling GPUs :)
docker run --gpus all -it -p 8888:8888 -v $PWD:/data tf_practice_gpu

Docker run arguments:

• --gpus all, enable all the gpus available
• -it, run the container as interactive mode
• -p 8888:8888, expose the ports to run jupyter from the browser
• -v $PWD:/data, mount your current directory ($PWD), into /data volume

Starting your jupyter notebook

Boom 💥 , now you can start running jupyter locally.

Go to http://localhost:8888