PyData # 28 - Explainability and Causal Inference


Details
Join us for our 28th PyData Meetup!
This time we have two great talks on advanced topics every Data Scientist would love to learn!
Agenda:
18:00 Mingling and snacks
18:30 Gathering and opening words
18:40 SHAP Values for ML Explainability: Intuition and Real-Life Examples – Adi Watzman
19:20 Short break
19:30 Introduction to causal inference in time series data - Shay Palachy
20:15 Some more snacks and mingling.
__________________________
See you soon, and thanks PayPal for hosting this event! :D
__________________________
All talks will be in English, and later uploaded to the PyData Youtube Channel.
__________________________
SHAP Values for ML Explainability: Intuition and Real-Life Examples – Adi Watzman
How do ML models use their features to make predictions?
SHAP opens up the ML black box by providing feature attributions for every prediction of every model. Being a relatively new method (arxiv.org/abs/1705.07874) , SHAP is gaining popularity extremely quickly thanks to its user-friendly API and theoretical guarantees.
In this talk I will guide your intuition through the exciting theory SHAP is based on, and demonstrate how SHAP values can be aggregated to understand model behavior. Throughout the talk I will present real-life examples for using SHAP in the fraud detection domain at PayPal, and in the medical domain as provided by the SHAP authors’.
About the speaker:
Adi is a data scientist at PayPal, developing Machine Learning models for fraud detection. She also volunteers in The public Knowledge Workshop (הסדנה לידע ציבורי), applying data science to improve public transportation in Israel. Adi loves data and algorithms, and is extremely excited about explaining machine learning models.
__________________________
Introduction to causal inference in time series data - Shay Palachy
In this talk I will give concise review of the major approaches found in academic literature and online resources for the purpose of inferring and detecting causality in time series data.
I will start with motivation, explaining why detecting causality is important, the many different use cases it has, and why it cannot be done intuitively (correlation does not imply causation). I will then briefly go over the main theoretical approaches suggested over the years to define causality, highlighting the way they differ and the impact they have. Moving on, I will present the prominent approaches to infer causality, born of the previous definitions, focusing on limitations and pitfalls and almost always referring to Python or R implementations of each approach. Finally, I will give a short guide to which approach to choose, depending on your data, research question, possible assumptions and KPIs.
About the speaker:
I ❤️ learning, data science-ing and making open source Python. I've founded the NLPH initiative and co-founded the ML-centric hackathon DataHack and DataTalks meetup series. I work as a data science consultant.
__________________________


PyData # 28 - Explainability and Causal Inference