PyData TLV meetup #2

This time we want to hear you guys!

We want to get to know you, your goals, your teams, your projects, the machine learning models you adore, the Python packages you like to use, and anything (Data related) you have to tell us!

We will do that by introducing a special kind of meetup called Lightning Talks.

A lightning talk is a very short presentation lasting only a few minutes. Several lightning talks will usually be delivered by different speakers in a single session, sometimes called a data blitz.

Here is the schedule for this special Lightning Talks meetup:

• 18:00 - 18:30 - Gathering, snacks, mingling

• 18:30 - 18:35 - Opening words

• 18:35 - 19:25 - Lightning talks - part I

• 19:25 - 19:35 - Break.

• 19:35 - 20:25 - Lightning talks - part II

The List of Lightning talks: (Each talk will take up to 10 minutes)

1. Uri Goren, Research Engineer at Yahoo

Basic Bayesian statistics for churn analysis

In this talk we would explore dwell time analysis (how much time a user spends on a website before churning) using the classic Frequentist method, and a simple Bayesian model that yields a surprising, yet very practical result.

2. Ohad Zadok, Chief Data scientist at EverCompliant

Visualization of Neural networks

Deep learning has made possible some of the recent breakthrough in vision and NLP domains. Training a neural network can have magical results but the way to achieve a good convergence is sometimes full with trial and errors. We suggest a visual way to understand the data manipulation layer by layer by using a PCA and transfer learning to visualize outputs from each layer.

3. Tal perry, data scientist at Citi

Analyzing financial chat using neural entity extractor

Billions of dollars of stocks and bonds are traded everyday over chat. Financial chat is just as messy as the chats we see on Twitter, WhatsApp or facebook, with emojis, slang and spelling mistakes making traditional methods of analysis less effective. In this talk, Tal will cover how Citi leverages the PyData stack and Tensorflow to build a neural entity extractor which has empowered the organisation to analyze that data.

4. Nathaniel Shimoni, Forecasting Manager at Strauss

Model validation methods (Kaggle)

In every predictive model we need to validate our results. This is done to make sure that we best fit our model to the train set while generalizing to unseen data as well. There are various ways to perform validation. In the talk I'll present some of the approaches I've gathered from my experience on kaggle.com (http://kaggle.com/) And the way to best match the validation method to the problem at hand.

5. Raphael Cohen, PhD. Senior Innovation Scientist @ Chorus.ai

Wow, that's awesome! Detecting emotion in call recordings.

Sales and Customer Success interactions with customers are a huge asset for a company. Conversation analytics, like Chorus.ai, allow customer facing representatives to easily record their calls and access the valuable information within them. Sales and product managers want to know which feature or selling point made the customer say “oh my gosh, that’s exactly what we wanted”. Looking only at the text generated using Automatic Speech Recognition (ASR) is not always enough (a sleepy “yeah, great…” is not the same as “Yeah! Great!”). I will present our python based method for ML identification of excitement and the research process of developing such tools.

6. Shlomo Kashani, Chief Data Scientist at ECI Telecom

10 minutes on Bayesian Machine Learning in Python.
In Bayesian analysis, the posterior follows from the data and a choice of a prior and a likelihood. This talk will discuss Bayesian Learning of the posterior and Binary Classifiers using PyMC3 and Stan for Kaggle competitions. Using the case in point of Bayesian logistic regression, this session will stimulate and outline the use of PyMC3 and Stan, two advanced software modeling tools for the simulation and the inference of Bayesian probabilistic models. Following the session fully working Jupyter Python notebooks will be available on Github.

7. Omer Sagy, data scientist at Dell

Sugar cane yield Prediction

The talk will present a solution that was developed by Dell’s Data Science as a service team in the agriculture industry. The developed tool calculates the predicted yield (in terms of tons per acre) of a given sugar canes field based on internal data sources such as soil type, fertilizers, plant ages, etc. and external public sources like satellite images and historical weather data. In a three weeks sprint, we delivered a model that is more accurate than the existing model while reducing the runtime from 48 hours to 20 minutes. The POC was implemented over Apache Spark and Python.

8. Amit Aronovitch, Algorithm Developer at BrightSource Energy

Quick visualization tool for Analysis of Shadows cast by (actual physical) Clouds

In the Solar Energy industry, understanding and predicting the patterns of light and shadow that are cast on the ground during cloudy days can be worth LOTS of money.
I will explain the context, so you can see why this statement makes sense, then show how basic capabilities of Matplotlib were used to "add interactivity" to
existing Python code, originaly written for benchmarking predictors for these patterns, and thus effectively converting the code into a visualization app that can actually be used by other people for case studies (and without writing any GUI-toolkit-specific code).

9. Daniel Fisher Levy, data scientist at MyHeritage

The Feather format

The Feather format is a fast, lightweight, and easy-to-use binary file format for storing data frames.

One of its best features is that it is language agnostic. In this talk we'll explore some ideas of why and how to use it for data cross-language data analysis in Julia, R and Python.

10. Gidi Shperber, Independent data scientist

Modern recommendation engine in python with LightFM

lightFM is a simple,fast, modern package for creating recommendation system in python.
in this talk, I'll present this package, along with some (very light) theory about recommendation systems, and the progress in this field since the well known netflix competition in 2009

11. Alon Nir, Data Scientist at HoneyBook

Pandas tips and tricks

Pandas is a powerful yet friendly Python library for data exploration, manipulation and analysis. In his talk Alon will share a few nifty tips and tricks to showcase Pandas' depth and breadth. Time permitting we'll also look at what's new in v0.19 and what's to come in v0.20.

Thank you all the speakers for agreeing to share your experience and ideas with us!

To understand what it would look like you can see how it went in PyData London 2015:

https://www.youtube.com/watch?v=i527Be2AnV8

Note that this is NOT a beginner's meetup, as all of PyData meetups we address professionals in the field to share ideas and get inspiration from one another! :)

The event is sponsored by Chorus.ai.