Machine Learning Engineer | Data Engineer | Data and AI Team@BT | Ex-TCS | Python | MS in Data Science
Recently I have gone through Hyperparamter tuning ability on GCP. I have explored two flavours of HPT namely Vertex HPT and Ray HPT. Though both are quite usefull , I have noticed some difference between them. Vertex HPT is managed HPT services from google cloud. It is quite powerfull to run 100s of concurrent threads. It is easy to use and integrates seamlessly with other GCP services. On the downside, its support for search space algorithm are limited. Ray HPT is built on top of Ray Framework. It is quite flexible and provides more control to implementor. On bright side, it supports wide range of search algorithms. Also, it is not limited to google cloud but can be implemneted on kubernetes clusters also. so, when to use what, Go for RAY HPT when you need more control(if not on cloud, comfortable in setting and managing your own cluster) while Vertex HPT will be saviour if you want quick and easy solution(minimum infra management). #MachineLearning #GCP #VertexAi #DataScience #Data #HPT #rayhpt #ray
