Automated data analysis and text narratives

Submitted Aug 31, 2017

Technical level: Intermediate

Autolysis is a domain-agnostic solution that solves pattern-based behavioral problems. At its core, Autolysis uses Groupmeans – a statistical technique that is designed to identify the impact of groups over metrics or vice-versa. An open source version with a friendly license is hosted at https://github.com/gramener/autolysis and a web interface is available upon request. Text narratives that augment the statistical analysis forms a key aspect of interpreting results in a business context. I will discuss the challenges in setting up the technology stack and building a narrative engine using templates and retaining necessary context.

Outline

Problem

What do these questions have in common?
1) Which of my channels is fetching me the most viewership? – Media
2) Which segment of customer will most likely want to buy expensive watches? – Retail
3) How is the customer demographics impacting the loan repayment? – Banks

All the questions fall under a similar pattern of questions: what is the impact of a metric on a group or vice-versa. Autolysis is a domain-agnostic solution Gramener built to solve pattern-based behavioral problems and where minimal human intervention is necessary to interpret the analytical results.

Technology stack

I will discuss the technology stack setup behind Autolysis web version which is built using Gramex – a Gramener data and web server which is stacked on python – and web sockets.

Text Narratives

Text narratives which retain insights, context, statistical relevance and augment the data analysis are an important aspect of explaining results. I will discuss the challenges involved in automating data analysis and smart text narration.

Requirements

No hardware/machines required

Speaker bio

I am interested in how automation impacts our daily lives. I work at Gramener as data scientist post my PhD at the Virginia Commonwealth University and a short post-doc at the University of Wisconsin-Madison.

I am interested in civic technology, automated learning, mapping for humanity, data journalism.

Comments

Previous Next

Iterators, Generators and You

Iterators in Python aren’t a matter of type but of the protocol i.e. any class that implements this protocol can be iterated over. This talk will present an overview of the all-encompassing iteration features in Python and how the Iterator pattern is built into the Python language so you never need to implement it by hand. more

31 Aug 2017

PyConf Hyderabad 2017