SC2018

Super Computing 2018 was one amazing experience and here's something I really loved doing there!

Introduction of Practical Approaches to Data Analysis for HPC with Spark

This tutorial was presented by The University of Tennessee, Knoxville

• Started with how to use the Jupyter Notebook
• They gave access to the Cluster at Oak Ridge National Laboratory
• Explained how to run code in cells
• Introduction to MarkDown
• Hands-on with Spark by running on the cluster
• Carried out k-means clustering
• Worked on DBSCAN
• Spark is used to define distributed datasets and implement any kind of operations
• The main purpose was to clean, transform, analyse large datasets with Spark on cluster using Cloud
• Concepts related to RDD (resilient distributed systems)

The two main concepts in Spark

Transformations

Actions

ReduceByKey is a special case where elements are distributed that have same keys.

First we create distributed datasets Second, we apply actions on this distributed datasets

Worked on NHANES Dietary dataset

This depicts the 48-hour diet plan of an American. This dataset will be updated on monthly basis and is high-dimensional

Problems addressed:

How to deal with missing data?

How to perform K-means, DBSCAN using Spark?

Note: Spark is fast because it uses the cache memory