Neural Networks - Python and Julia

Objective

These are a Python and Julia implementations of a neural network for performing training and prediction under common specifications defined for multiple languages.

The objective of this project is to benchmark both languages on a common task, and to compare the performance of the neural network implementations in both languages.

Deployment

Requirements

• Docker >3 (needs docker swarm)
• Julia (for local builds)
• Python 3.10 (for local builds)

Configuration

• Dataset: in data/archive.zip you can find the dataset used for training and testing the neural network.

Commands

Startup

• make build will build the docker image used in both languages.

Run

• make run will run the system. The notebooks will be available at http://localhost:8888.
• make remove removes all services.
• make datasets will split the dataset into training, testing and validation sets.

Additionally, the notebook training_pytorch.ipynb must be executed by a Python 3.10 interactive kernel which should be already available into the container.

Monitoring

• Grafana: http://127.0.0.1:8081
• Graphite: http://127.0.0.1:8080
• Logs

Libraries

• PyTorch
• Tensorflow
• Scikeras
• ScikitLearn
• Flux.jl
• Pandas
• Numpy
• Matplotlib