This is a demo to show you how to run a handwritten digit recognition model on Clusterone. The demo uses TensorFlow and the MNIST dataset.

Follow the instructions below to run the model on Clusterone using the just command line tool. This project is part of Clusterone's Getting Started guide. There is also an in-depth tutorial based on this repository.

Please note: There is currently a bug in the MNIST example that might affect the performance in distributed training. We are aware of this and are working on a fix. This does not affect single node training.

• Install
• Usage
• More Info
• License

To run this project on the Clusterone platform, you need:

• Python 3.5
• Git
• The Clusterone Python library. Install it with pip install clusterone
• A Clusterone account. Join the waitlist if you don't have an account yet.

Start out by cloning this repository onto your local machine.

git clone https://github.com/clusterone/mnist

Now you're all set to run MNIST on Clusterone!

These instructions use the just command line tool. It comes with the Clusterone Python library and is installed automatically with the library.

cd into the folder you just downloaded with cd mnist and log into your Clusterone account using just login.

First, create a new project on Clusterone:

just init project mnist

Then, upload the code to the new project:

git push clusterone master

Finally, create a job. Make sure to replace YOUR_USERNAME with your username.

just create job distributed --project YOUR_USERNAME/mnist --module mnist --name first-job \
--time-limit 1h

Now all that's left to do is starting the job:

just start job -p mnist/first-job

That's it! You can monitor its progress on the command line using just get events. More elaborate monitoring is available on the Matrix, Clusterone's graphical web interface.

For further information on this example, take a look at the tutorial based on this repository in the Clusterone Documentation.

For further info on the MNIST dataset, check out Yann LeCun's page about it. To learn more about TensorFlow and Deep Learning in general, take a look at the TensorFlow website.

MIT © Clusterone Inc.

The MNIST dataset has been created and curated by Corinna Cortes, Christopher J.C. Burges, and Yann LeCun.