[研究] 關於 Bayesian Global Optimization

目前我們使用的是Python 版的 gpyopt

另外有商業版的：SigOpt 

介紹關於Bayesian Global Optimization：

• https://www.infoq.com/presentations/bayesian-optimization-ml
• Bayesian Optimization for Hyperparameter Tuning
• Automatic Features Generation And Model Training On Spark: A Bayesian Approach
• 
  

一般常用的兩種方法是，但是都很吃資源：     

• Grid search     
• Random search

關於Spark 有支援的lib：

• https://github.com/deeplearning4j/deeplearning4j/tree/master/arbiter
• http://spark.apache.org/docs/latest/ml-tuning.html

其它關於SigOpt 討論 Bayesian Global Optimization

• https://news.ycombinator.com/item?id=12257579

---

目前市面上有的選擇：Optimizing hyperparams with hyperopt

• Python
  

  https://github.com/automl/SMAC3  

  https://automl.github.io/SMAC3/stable/manual.html

  http://hyperopt.github.io/hyperopt/ (Distributed Asynchronous Hyperparameter Optimization in Python) 
• Java
  

   http://www.cs.ubc.ca/labs/beta/Projects/SMAC/

---

介紹 Bayesian Global Optimization 的投影片

• https://www.cse.wustl.edu/~garnett/cse515t/spring_2015/files/lecture_notes/12.pdf
• http://gpss.cc/gpss15/talks/gpgo-gpss2015.pdf
• http://gpss.cc/gpmc17/slides/LancasterMasterclass_1.pdf

關於 Bayesian Optimization 的應用

• Automated Performance Tuning with Bayesian Optimization
• Continuous Optimization of Microservices Using ML
• BOAT: Building auto-tuners with structured Bayesian optimization
• 
  

---