Recent Developments in LightGBM

Recent Developments in LightGBM

Update: Video recording:

Slides and code: here.


After our last meetup with an overview and benchmark of the most popular GBM implementations and then a separate talk on XGBoost, we are back with a talk about LightGBM this time, again right from one of its core developers and maintainers. This time we’ll fit in a 1-hour slot, talk 35 minutes + Q&A 20 minutes (10:00-10:55am Pacific Time).

Recent Developments in LightGBM

by James Lamb

In this talk, attendees will learn about recent developments in LightGBM, a popular open source gradient boosting library. The talk’s primary goal is to educate attendees about recent work that has been done to make LightGBM easier to install and use effectively.

The talk will describe in detail a few recent additions to the library, including:
* CUDA-based GPU acceleration
* new options to make installation of a GPU-enabled Python package easier
* availability of the R package on CRAN
* new detailed documentation on hyperparameter tuning
* Dask integration for distributed training on large datasets

Demos of the new (experimental) Dask integration will be shown. The talk will conclude with a summary of new features that are coming in the future, and a call for attendees who are interested to contribute their ideas, documentation, and code to LightGBM.

Speaker Bio:

James Lamb is a software engineer at Saturn Cloud, where he works on a managed data science platform built on Dask. Before Saturn Cloud, James worked on industrial internet of things (IIoT) problems as a data scientist at AWS and Chicago-based Uptake. He is a core maintainer on LightGBM, and has contributed on other open source data science and data engineering projects such as XGBoost and prefect. James holds Masters degrees in Applied Economics (Marquette University) and Data Science (University of California, Berkeley).

Date/Time: Tuesday, January 12, 10:00-10:55am Pacific Time
Venue: online (zoom)
RSVP: here on meetup
Note: The zoom link will be posted in comments on meetup (link above) at 9:55am and due to our zoom’s 100-attendee limit, the first 100 people will be able to join the zoom call.

Don’t miss the chance to interact with a core developer of one of the most popular GBM libraries.

Share This Post

1 Comment

  1. Scott Edwards - January 12, 2021

    Congrats on winning that bet, Szilard! This was a great talk, James! I’ve been enjoying exploring Dask on Saturn Cloud – it’s a great alternative to Spark. RAPIDS AI is interesting too — and has Dask as well — but not pure open source since NVIDIA is proprietary code.

Leave a reply