Scaling Your Data Science Experiments – from Jupyter notebooks to 6,000 GPUs

Abstract: Data Scientists and Machine Learning professionals today face a quandary of choices when trying to figure out how to scale their data science experiments. What tools should I use (Keras, TensorFlow, Scikit-learn, PyTorch, etc.)? What are the best ways to share my models with customers / partners? How do I manage cost efficiently? and most importantly how do I scale my project to a full blown production? This presentation will give attendees the information they need to better understand the landscape of options available to them, give them helpful suggestions on what routes work and which don't and how to make best use of the free and open source tools available.

This presentation will cover:
- Setting up the best ML test environment for the problem you're trying to solve
- How to run experiments locally, remotely or in the Cloud
- Top tips for monitoring and organizing experiments
- Cost efficient way to spin up compute
- Best model libraries to use and for what
- Best ways to manage artifacts and perform hyperparameter search.