Official account for CatBoost, @yandexcom's open-source gradient boosting library https://teapawt.net/t.co/LWqilHFELV
Moscow, Russia
Joined August 2017
- Tweets303
- Following38
- Followers2.3K
- Likes125
#catboost_tipsntricks CatBoost sets a learning rate by looking at the number of iterations&objects in the trainset. In today's video, Nikita explains how to use built-in interactive learning curves to tune LR & iterations and improve model performance.
youtu.be/O2OJ_JWYV0I
#catboost_tipsntricks Consoles are not only for Jupyter&Python😸 In today's video Kate explains how to use main CatBoost features from CLI. This simple but powerful interface allows you to use practically anywhere and improve ml pipelines. youtu.be/m3E35snIrAM
CatBoostML retweeted
Gradient boosting methods have been proven to be an important strategy. This article with @neptune_ai aims to investigate and compare the efficiency of three gradient methods focusing primarily on @CatBoostML. bit.ly/3fGFSjS
#catboost_tipsntricks 📹Model prediction interpretation in a human-readable form is a key for making a great machine learning system. In this video Nikita shows how to use SHAP values to understand model predictions youtu.be/RNT1o2gu5Ms
Technical notice⚠️ In the next release, we will stop publishing CatBoost artifacts for Python 2.7 & 3.5 versions. If you still need CatBoost built for 2.7 or 3.5 - you can build it from sources.
If you have any questions - contact us here, in telegram or via GitHub issues!😺
#catboost_tipsntricks Feature selection is a crucial part of data engineering & ML. In today's video, Ivan talks about CatBoost's built-in feature selection function. It can help you speed up training and reduce overfitting.🚀 youtu.be/iuRGv31mcuI
#catboost_tipsntricks If you use GBDT models in production, don't miss that video😺 Ekaterina Ermishkina explains how to apply CatBoost models in different formats and environments: native binary format, CoreML, PMML, ONNX, in Java, Rust, NodeJS and others youtu.be/fpUEoy60x24
#catboost_tipsntricks In today's video, Nikita Dmitriev talks about object importance and how you can use it to detect and drop noise objects and boost the quality of your models🚀Stay tuned for the next episode! 😺 youtu.be/ce1VULptNWQ
We've recorded a series of short videos to boost your CatBoost knowledge, so stay tuned😺
In today's video, Ivan Lyzhin explains why you should try different tree grow policies.
youtube.com/watch?v=lhaOYwOx…
🚀🎇1.0.0😺🥳
It's not only CatBoost's 100₂ anniversary but also a first major release! We upgraded the major version because CatBoost looks solid: by the last four years, CatBoost began to play a crucial role in Yandex, CERN, and many other companies.
And let us introduce the main features: fully distributed training for Apache Spark, multilabel multiclassification, big CPU training time speedup (up to 35%), improved CV speed, LogCosh loss, model size regularization fix for GPU, markdown documentation. And that's not all!
See full release notes in our github github.com/catboost/catboost… 😺🥳
🎇We switched main url to new documentation! Old documentation would be available at catboost.ai/docs-old/ for next two weeks. If you'll find some problems with new documentation and will need old docs available - contact us here or in telegram t.me/catboost_en 🐱
We'd like to invite Russian-speaking followers to our 100₂th online birthday party. Read more and register now: events.yandex.ru/events/catb…
And don't worry! We are planning to translate the recorded video into English and publish links later here. Stay tuned! 😸
Good news, everyone! We've refactored CatBoost documentation and are inviting you to test it here: catboost.ai/en/docs-beta/ And from now on documentation sources in Yandex Flavored Markdown can be easily found in our repo
github.com/catboost/catboost… We are waiting for you PRs!😺
#CatBoostPoll CatBoost already supports distributed training on Apache Spark and by separate processes from CLI. If you'd like CatBoost to support your favourite framework - please vote or reply with your variant😺
47%Dask
12%Ray
35%Spark
6%MPI
17 votes • Final results CatBoostML retweeted
New paper: we tested how different ML methods perform on predicting administrative errors in US unemployment insurance data. Turns out: @CatBoostML is more accurate, along several measures, than every deep learning model tested. (open access for two weeks) dl.acm.org/doi/10.1145/34636…
🚀CatBoost 0.26.1 😺
News:
* R package: supported text features and virtual ensembles prediction
* New MultiRMSEWithMissingValues loss function that supports training multidimensional regression models with missing labels github.com/catboost/catboost…
We published the second part of our epic uncertainty saga: Estimating Uncertainty with CatBoost Classifiers by Andrey Malinin in @TDataScience
We show how uncertainty allows to detect samples that are unknown to model on KDD intrusion detection dataset 😺 towardsdatascience.com/estim…