group2-machine-learning
Introduction
Project phases
- Making the repository and the required materials.
- Build docker containers for the MNIST analyses
- Use workflow tools + cloud/HPC computing to run MNIST analyses
- Repeat steps above for fashion MNIST and humback whale classification if time.
Instructions
Initiate a virtual machine through a cloud server
- Log in to Jetstream and click "Start a New Instance."
- Select the Ubuntu 18.04 Devel and Docker instance and press launch.
- Select an m1.medium instance in instance size, and then click "Launch Instance."
- Once the instance is "Active," go into the shell either through the Web Shell or via ssh.
Run the docker image
-
Pull the docker image with the command below. You can use
sh docker imagesto check if you successfully pulled the image.docker pull username/machinelearning -
Run the image with the command below.
docker run -it username/machinelearning shThe default is to run a neural net classifier on all of the MNIST, fashion MNIST, and Humpback Whale Tail datasets. If you would like to specify a dataset, you can add it with the
-etag (see example below). Options are mnist, fashion, whale.docker run -e dataset=mnist, -it username/machinelearning sh -
The neural network builder and classifier has launched! When it is finished, you will see an output file with a text and pdf summary of the results. Compare your results to the example output below
Further exploration
MNIST classifier with mlp neural network
To work with the neural network model, press the button below to launch a jupyter notebook (Note: this notebook runs on Binder servers).
