Hacking Kaggle Challenges

Git:

https://github.com/QuantScientist/deep-ml-meetups/tree/master/hacking-kaggle

Location:

3rd Fl. J&J Meeting room, Elevator entrance behind Aroma Cafe.

Kaggle #1 : Hacking Kaggle Challenges

Dear Prospective Kagglers,

All levels from complete novices to experienced data scientists are welcome at the meetup. During this meetup you will:
· Take part in a Kaggle competition, team-up with other challengers for future competisions.
· Challenge your techniques, and learn new ones … even those not listed in books ;)
· listen, learn, pitch, or … code in a friendly mood
· Discover inspiring starter codes

All in all, a productive evening !

Tentative Schedule:

• 17:30 Doors open, get a drink and meet cool data people.
• 18:00 Full Jupyter/Python pipeline for the NIH challenge -Shlomo Kashani.
• 19:00 Starting Data Science with Kaggle - Nathaniel Shimoni

• 20:00 Click Prediction, Gidi Shperber

(1) Predict seizures in long-term human intracranial EEG recordings.:

https://www.kaggle.com/c/melbourne-university-seizure-prediction

This is a binary classification problem. Therefore, using Jupyter, I will explore several algorithms including:

• Feature generation based on EEG signals

• Logistic Regression (sk-learn)

• Bayesian Logistic Regression (PyMC3)

The NIH contest is doable for beginners and challenging for experts. Everyone will find their share.

(2) Starting Data Science with Kaggle:

To start off I will try to motivate you what the advantages are of doing Kaggle (compared to other activities) and how it may help you to overcome the dichotomy connected to the being of a Data Scientist.

After you got motivated I show you a setup for your Kaggle projects to make your Python data science projects isolated, reproducible and well structured to enable you to work with others without much fuzz.

If you want to see what competitions are currently available, check them out here (https://www.kaggle.com/competitions). If you are a complete novice check out the Titanic tutorials.

(3) Click Prediction

Click prediction competitions appear less on Kaggle then image recognition and other deep learning related subjects. However, click prediction is the main task of data scientists in organizations. And unlike other sub-fields, companies don’t hurry to share their models, insights and definitely not the data. Therefore, these competitions are extremely valuable.

In this lecture we are going to discuss the Outbrain click prediction competition - the data, algorithms, and general kaggle tips, tricks and best practices.

In the end of the lecture you’ll be able to successfully compete in such a competition, or have a good start for implementing such a solution in your company.

Bios:

Nathaniel Shimoni – Kaggler & Data science evangelist.
Tackled more than 35 different challenges @ kaggle.com
Currently ranked 199 of over 50,000 active Kaggle users
Holds BA in economics & MBA with emphasis on entrepreneurial studies both from Ben Gurion University.
Was the team leader of “Sea through data” 1st place winner on windward challenge in Data-Hack 2016 hackathon. Works as Forecasting Manager @ Strauss.

Gidi Shperber: is a freelance machine learning engineer, helps startups to build machine learning oriented products, Especially in fields of computer vision and personalization.

Github code:

https://github.com/QuantScientist/deep-ml-meetups