Snap ML: Accelerated, Accurate,Efficient,Machine Learning
summary
Library codebase is C++/CUDA with Python wrappers following the scikit-learn API, offering CPU, multi-threaded, GPU, multi-GPU and distributed solvers and an inference engine that imports XGBoost/LightGBM models. Benchmarks claim typically 10x training speedups on CPU and/or GPU, a boosting variant SnapBoost that after hyperparameter tuning beats XGBoost and LightGBM on 32 of 48 OpenML datasets, and inference latency gains up to 32x for single-record scoring on IBM Z, reported to hold while matching accuracy.
AI-written summary · about
chapters (14)
- 0:00Introduction
- 3:33Snap ML overview
- 5:20Research innovations
- 8:54API and examples
- 12:27Logistic regression benchmarks
- 15:00Decision tree benchmarks
- 16:05Random forest benchmarks
- 17:08GPU vs RAPIDS
- 18:32SnapBoost accuracy
- 20:39SnapBoost on Kaggle
- 23:30Industrial SnapBoost
- 24:33Inference acceleration
- 28:11Publications and resources
- 28:32Install and try
related talks
42mSnap ML: a highly-accelerated, scalable software library for machine learning
32mTaming the Machine: Basics of ML Models Training and Inference Optimization
16mAmeet Talwalkar, Assistant Professor of Computer Science, UCLA @ MLconf SF Part 2
10mMachine Learning for mobile
23mHugging Face Accelerate: Making Device-Agnostic ML Training and Inference Easy...
21mKeynote: Easier Machine Learning Thoughts From Scikit-Learn
34mSpeeding up Machine Learning tasks using GPUs in Python
26mNext generation machine learning for mobile and embedded platforms