在实际工作中，data science和machine learning 承担的任务有什么不一样

这个问题有一部分原因是各个公司job title的混乱，但是抛开title，两个职位所用技术有什么不同，面临的任务有什么不同呢？

也就回答过1000 多次吧……

我在fb做ds，刚进来的时候ic3，有点挣扎，经常加班，偶尔写写r和pipeline (python)，用的最多的还是sql，一年挣十几万。现在ic5了，几乎连sql也不写了，弄几个chart搞个analysis到处去讲故事去Influence，meet expectation还算轻松，不怎么需要加班，一年挣二十几万。这楼里提到的前沿技术基本都没接触过，不知道再过几年自己是否还能survive :(
我的case不能说典型，毕竟公司很大，ds里什么样的archtype都有。

ds像data analyst的升级版，以分析数据或者搭建分析平台为主，代码一般不直接进产品。
mle更多是backend中间的middle layer，不仅需要train model 还要考虑如何serve model (比较难的是real time serving, 简单一点的是offline生成结果直接被consume或者存在store里serve。
更粗暴的说，mle的工作主要集中在user request来和返回response的过程之间，ds则是在这一过程之后进行分析。
当然这种分类不绝对的，也有公司管mle叫ds 或者mle搭建数据监测分析平台的。

好问题啊，顶一下，期待在职人士回答

ds->modeling
mle->engineering

wryswa 发表于 2020-1-16 10:38
ds像data analyst的升级版，以分析数据或者搭建分析平台为主，代码一般不直接进产品。
mle更多是backend中 ...

多谢多谢，看来还是mle更是技术活

ricaelum42 发表于 2020-1-16 10:59
ds->modeling
mle->engineering

说的有理

K姐 发表于 2020-1-16 13:33
也就回答过1000 多次吧……

来了大神呀，再追问一句，准备ds和mle的面试有什么区别，感觉大家都在刷题和上ml的公开课，那用的技术有区别么

Depends.

But title is not the key here. The same people can be titled as DS, MLE, ML scientist, or even research scientist. But very likely a DS are required to undertake some data integration/cleaning work or even ELT work (like data engineer), but it is not common for MLE. A data scientist could be asked to develop backend for products, like MLE, and I see many cases.

MLE DOESN'T have to do any ml modeling/training work sometimes, BUT they have to wrap the model to the product pipelines.

What makes a DS very competitive from my eyes: 1. experience in cloud tools, e.g., AWS sagemakers/athena/EMR 2. Strong theory background in DL. I asked people to explain what BERT truly does and why dropout can't be used before BN, 80% of them can't give me good hits. 3. Stop messing with SQL, fu@ck it, learn to use Spark  (which means knowing JAVA makes you very outstanding for true big data driven positions).

What makes an MLE very competitive, coding and lot of product development experience. Anyway many sde jobs are titled as MLE.

No Chinese input here, sorry.

https://www.youtube.com/watch?v=xC-c7E5PK0Y