AI Playground provides an easy way to setup a tensorflow, scikit-learn and jupyter notebook playground as a docker container environment. My goal is to help you easily start learning ML and AI basics without spending a few hours to setup the environment.
I assume that you already know the basics of Docker such as docker build and run. Thanks to Google, a good base image is available at Docker Hub: Tensorflow repository. You can simply start from a Google's tensorflow image as the base docker image and add layers on top of it as you need. For example, I included scikit-learn version 2.0-dev package instead of the version 0.19x stable since the 0.20-dev version has many bug fixes and new features that we need to run sample codes from a very useful reference, Hands-on Machine Learning with Scikit-Learn and Tensorflow If you haven't read this book, I recommend getting a copy.
First, login to your laptop or PC. I assume your laptop is either Linux desktop or mac. You can use Windows as well but I haven't tested it yet. If you haven't installed Docker on your PC, install Docker first by following instructions at Docker Installation
Once you have Docker running on your laptop, run following commands to prepare aiplayground docker image. I could have shared the docker image in a public repository but I decided rather not redistributing stuff without fully validating licenses. You can easily build your docker image by following instructions below anyways.
cd
mkdir -p projects
cd projects
git clone https://github.com/erickang7/ai
cd ai/tensorflow/docker
./build.sh
docker images | grep aiplaygroundaiplayground/tensorflow:practice image should be created.
Let's spin up a new container and call it as tensorflow. Simply run the following command.
docker run \
--name tensorflow \
-v $HOME/projects/ai/practice:/notebooks/workspace \
--mount source=tfvol,target=/notebooks/models \
-p 8888:8888 \
-p 6006:6006 \
-d aiplayground/tensorflow:practiceThen run the following command to get the endpoint URL to yupyter notebook running in the container.
docker logs tensorflow | grep tokenYou should see a result similar to the sample output below. In the sample output, http://localhost:8888/?token=83a3719cec38fb6849d708fb757d5adec1f72b159f73b172 is the URL for the jupyter notebook.
[erickang@bluewood tensorflow]$ ./init.sh
0a6089b5b96d24de5bb8faa2027b0e2b5d23c61f2b784469db0206e1cacc1185
[I 20:50:51.681 NotebookApp] Writing notebook server cookie secret to /root/.local/share/jupyter/runtime/notebook_cookie_secret
[W 20:50:51.695 NotebookApp] WARNING: The notebook server is listening on all IP addresses and not using encryption. This is not recommended.
[I 20:50:51.701 NotebookApp] Serving notebooks from local directory: /notebooks
[I 20:50:51.701 NotebookApp] 0 active kernels
[I 20:50:51.701 NotebookApp] The Jupyter Notebook is running at:
[I 20:50:51.701 NotebookApp] http://[all ip addresses on your system]:8888/?token=83a3719cec38fb6849d708fb757d5adec1f72b159f73b172
[I 20:50:51.701 NotebookApp] Use Control-C to stop this server and shut down all kernels (twice to skip confirmation).
[C 20:50:51.702 NotebookApp]
Copy/paste this URL into your browser when you connect for the first time,
to login with a token:
http://localhost:8888/?token=83a3719cec38fb6849d708fb757d5adec1f72b159f73b172Also note that jupyter notebook is exposed at the port 8888. If you restart this docker container, you may need to get the new endpoint URL by calling docker logs tensorflow command again.
Open Jupyter notebook in your web browser by ctl+click the URL in terminal.
In the root file browser, open 1_hello_tensorflow.ipynb file. Then run the hellow world notebook to validate your environment.
For tensorflow basics, go to Getting Started for ML Beginner on Tensorflow tutorial site. You can use workspace/tensorflow/tensorflow_iris.ipynb notebook for the iris categorization tutorial.
For ML in general, Again, I recommend Hands-on Machine Learning with Scikit-Learn and Tensorflow. Read and practice it with sample codes in the book.
You can also play with data you already have such as telemtry data collected into a mssql db or auzre sqldb. To export all tables in a database to CSV files, you can use dbexport app written in golang. Check out dbexport git repository for the detail.
If you want to play with tensorflow samples which are written in python, you can do it by connecting to tensorflow container's bash terminal and using vim editor. To enter the container and test run the iris categorization DNN sample, run the following command.
docker exec -ti tensorflow bash
cp /gitclone_models.sh /notebooks
cd /notebooks
./gitclone_models.sh
cd /notebooks/models/samples/core/get_started
python premade_estimator.pyEnjoy!