Hyperparameter Optimization (HPO) with RAPIDS on Microsoft AzureML

This video tutorial walks through an example of accelerated hyperparameter optimization (HPO) jobs using RAPIDS on Microsoft AzureML. Each job uses RAPIDS dataframe library, cuDF, to load and process 20 million rows of airline arrival and departure data and build a model using RAPIDS machine learning library, cuML. Random Forest is used to predict whether or not a flight will arrive on time. AzureML trains and evaluates models with 100 variations of key parameters in order to find the combination that yields the highest accuracy. In this example, RAPIDS will reduce model training time by 25x, which will allow you to train dozens of models on the GPU, in the same time that it would take to train a model on the CPU.

Below are the three chapters of this tutorial:

1. Creating a resource group and workspace on AzureML
2. Setting up a local environment and launching a Jupyter notebook
3. Using the notebook example to launch a docker container on Azure for hyperparameter tuning with RAPIDS on GPUs

Jump in and give it a try. Find us on Slack or file a GitHub issue with suggestions.