PyConFr 2014

dimanche 15:00:00–15:30:00

Lean data science, or how to do machine learning with what you have to hand

Christophe Bourguignat

Audience level:: Intermediate

Description

That’s life : it’s not uncommon to have to do complex data science or machine learning on a capacity-limited platform : a few Gb of RAM, 2 or 3 cores only, no Hadoop Map / Reduce cluster, or an old laptop.

In these situations, you must do more with less, and do kind of “lean machine learning”.

Abstract

In this talk, we will show typical illustrations of such “worst-cases”, encountered in real life, or in data science challenges like Kaggle. We will give some tips & tricks, and best-practices to make the most of Python, and its flagship data analysis libraries : scikit-learn, pandas, numpy and scipy.

On the menu for this session : sparse matrix, incremental learning, chunk reads, manual caching, … and much more !