PyData TLV meetup #5

Our 5th PyData Tel Aviv event will bring us back to lightning talks format! We will include several 10 minute talks (similar to our 2nd event) on a variety of Python and Data Science topics.

Please note that the event will be in WeWork Ibn-Gabirol.

Thank you very much to Intuit Israel, who will be sponsoring the event.

The schedule is

• 18:00 - 18:30 - Schmoozing & noshes

• 18:30 - 19:20 - Welcome Word and First Session Of

• Simplifying Models with Model Compression

• Optimal distribution of taxis in new markets

• JamBot

• Near-real time detection of account takeovers in a login system

• 19:20 - 19:30 - Break

• 19:30 - 20:30 - Second Session of Talks

• Useful Decorators for Data science

• Data Science at Scale

• Cookie Cutter for Data Science

• EduTech

• Using Machine Learning in Small Business Lending

• Building fast and powerful models with very little data by augmentation method

As usual, all talks will be in English and filmed. Below are the abstracts.

Speaker: Monica Hsu, Intuit

Title: Near-real time detection of account takeovers in a login system

Abstract: Online login systems are often targets of account takeover attempts, in which fraudsters attempt to gain financially from personally identifiable information contained in customer accounts. Intuit Turbotax utilizes a multi-layered defense system to protect our customers from account takeover. In this presentation, we describe just one of these layers of defense: a system that combines near-real time feature generation and machine learning to detect and stymie account takeover attempts.

Speaker: Luis Voloch, Palantir Technologies

Title: Simplifying Models with Model Compression

Abstract: It is often possible to transform complex models into much simpler ones without significant loss of performance. In this talk we will cover the concept of model compression and show how it can be used to simplify models.

Speaker: Amir Livne Bar-On, Via

Title: Optimal distribution of taxis in new markets

Abstract: Via is a taxi booking app that uses the Sherut model: dynamically changing service lines and allocating vans, and asking the customer to go to a nearby street corner.

When expanding the service to a new city there's initially little demand but only a few vans to serve it, so it's crucial to spread them well.

I'll present the details, and an approach using a loss function, for which it is possible to find the global minimum in reasonable time.

Speaker: Diane Chang, Intuit

Title: Using Machine Learning in Small Business Lending

Abstract: Small businesses have a very difficult time getting access to the funding they need to help grow their businesses. At Intuit, we are trying to make the process much easier for our small business customers. One of the ways we do that is by helping them get benefit from their data that they have collected in our accounting software. With their permission, we utilize that data to predict when they may need financing, which lender(s) they are most likely to get approved by, and how likely they are to pay off the loan.

Speaker: Yigal Weinberger, Verint

Title: Data Science at Scale: Using dask for performane and scalability

Abstract: Today there is a great need for improved performance and scalability even in the early stages of the data science process, in my talk I will discuss the limitations of pandas in terms of performance and show a few hands-on examples for common data processing procedure using Dask

Speaker: Dalya Gartzman

Title: In EduTech we really are making the world a better place. And we are using Python to do it easily.

Abstract: Simplisico is an EduTech company bringing private math tutoring to the 21st century, utilizing DL+NLP to solve word math problems. In this talk I will use Simplisico's journey of joining the DL revolution as a case study for a process we will be seeing more and more of, where companies with no seemingly DS background will use "off-the-shelf" tools to integrate state of the art DS into their products.

Speaker: Uri Goren, Yahoo

Title: Useful Decorators for Data science

Abstract: Abstract oriented programming is a paradigm for minimizing cross-cutting concerns, and python supports it via decorators. We will cover 5 useful applications in the context of working within a Jupyter notebook.

Speaker: Tal Ben Yakar

Title: Building fast and powerful models with very little data by augmentation method

Abstract: Data are most crucial and essential building component for any data mining and AI applications exist. More significantly, deep learning approaches require massive datasets. We know that the theory and algorithms have been around for quite a while however the ability to process the right amounts of data brought us to the recent breakthroughs in the field.

A challenge comes up in a case of a small dataset, comparing to the required training data required. However, mostly, getting this data are neither an easy nor a cheap task, many annotating services take advantage of the problem and charge for tagging data-sets campaigns, those could cost hundreds of dollars easily and yet with an uncertain quality. as the task of generalization at hand, we wondered how to exploit the minimal data we have and still have an AI system to learn well. In this paper, we overview methods for solving the problem and suggest solutions in order to overcome the challenge.

Speaker: Hanan Shteingart, gong.io

Title: Cookie Cutter for Data Science

Abstract: A well-defined, standard project structure means that a newcomer can begin to understand an analysis without digging in to extensive documentation. It also means that they don't necessarily have to read 100% of the code before knowing where to look for very specific things.

Well organized code tends to be self-documenting in that the organization itself provides context for your code without much overhead. People will thank you for this because they can: collaborate more easily with you on this analysis, learn from your analysis about the process and the domain, and feel confident in the conclusions at which the analysis arrives.

Speaker: Tal Baumel, Ben Gurion University

Title: JamBot!

Abstract: A RNN based midi generator that you can play along with.