GPU Open Analytics Initiative

Accelerating the Scalable Data Science Environment with GPU-enabled Python

KDD'18 Hands-On Tutorial

Tuesday 8:30 am

---

Software / Hardware Requirements

The tutorial will leverage cloud resources that will provide the a common environment for all students.

Requirements:

• Laptop with WiFi

  • We will be using the conference WiFi, please ensure that you can connect prior to the tutorial

• Web browser - latest version of any will work, preference is towards Firefox or Chrome.

Tutorial Agenda

Introductions

• Who we are

Getting Connected

• Connect to Qwiklabs
• Introduction notebook to validate

Introduction and Background

• Big Data Ecosystem
• Challenges in Big Data today
• Apache Arrow
• GPUs for compute
• The GPU Open Analytics Initiative
• The GPU Data Frame (GDF)
• Python library for GDF (PyGDF)

Hands-on: Data Loading and Manipulation

• Lab 1: Data Loading and Manipulation

  • Traditional interface through Pandas
  • Pandas to/from PyGDF
  • Column Function and Basic Transforms
  • Filtering

• Student Assignment

Break

Hands-on: Data Science and Machine Learning

• Lab 3: Classification using XGBoost
  • Familarize with IoT cyber network data
  • Data ingest and feature extraction
  • Time binning and preparation for classifiation
  • Building XGBoost model
  • Evaluating the model via ROC curves and AUC
  • Student Assignment:
    • Investigation into other time binnings, aggregations, and XGBoost parameters
    • Using additional features (quantitative and categorical) in the data to build better models
    • Moving beyond connection logs to other log types (e.g., DNS) and building models

Break

Wrap-up and Conclusion-

• Roadmap
• Scaling out to multi-GPU and multi-node
• Partner Activities
• Comclusion