MLbase A Distributed Machine-learning System
1 minute read ∼ Filed in : A paper noteQuestions
This is mainly use some traditional techniques, and simply combine multiple ML stage into a pipeline.
Introduction
Background & Motivation
How to choose algorithms and how to scale the ML are challenging tasks for ML researchs.
Goal
The paper presents MLbase:
- Declarative way to specify ML tasks
- Optimizer to select and adapt the learning algorithm with a sophisticated cost-based model.
- high-level operators to enable ML researches to implement ML methods.
- Run-time optimized for data-acces pattern.
Some Optimizations used:
- use pruning heuristics and online model selection tools to improve the searching efficiency.
- Apply data process techniques to before attempting other more complicated techniques.