Data Scientist 炼成记录-更新完毕2018年12月 | 机器学习练成记录 - 已开新帖

Story telling, docs, slides etc.
Early on, spend a little bit of time to learn design (books and resources listed in the first floor)
each time you have the chance to write a doc/present something at work, stick with a consistent theme (so it is your signature theme, even if people don't consciously notice it), and tweak the design to make it pretty, and consistent.
It sounds like a waste of time, trust me, it's not....

After a while, this step wont' even take that long, and you will be able to whip out something that's scientifically sound (duh) but also aesthetically pleasing.

It also makes it that much harder for people to steal your work directly --- e.g. it's quite difficult to completely change the design of slides.

it's pretty amazing how fast things are changing nowadays
Now i look back at the study material and realize how much new tutorials/videos/MOOCs are now available
and so many packages would just "work" with a few lines of code
学习的门槛被dramatically 降低了
这也是身处一个“朝阳”或者“正午”产业的好处，欣欣向荣的发展和不断推陈出新，的确是让人兴奋
And looking back at the past 3-4 years, I feel so lucky to be in such a dynamic field, and to have learned from many world class scientists/researchers and worked along side top engineers/architects.

看着这个觉得心里有一点底，以前总觉得拳头打在棉花上，跟着Coursera上课上完以后也不过just so-so,好像自己学的都没有用。我是商科背景，好在我们本科学的计量经济学、运筹等还算是挺有用，刚刚到美国开始读DS的master.在学校里老师能教的东西太少了，然而看着这个帖子觉得自己还有好多要做，我这两年时间来得及么。我不想读phd了，也不知道脑子够用不。想知道K姐是什么经历呢。
关于国内的DS行业前景有什么看法呢，我想回国就业的。

请问，那本onlineststbook的练习，哪里有全部的答案？

同问国内的DS行业前景 谢谢

just found this fantastic site... https://see.stanford.edu/Course
it covers good content for basic coding (106A, B, 107), Andrew Ng's ML, and optimization
this would be SUPERB to add to the basic MS in analytics/ds/stats, and give a much better rounded education towards the full stack data scientist.

interviewed quite a few with MS in analytics, (Chinese and nonChinese),maybe even a couple of year's working experience, who want to be in DS (and not dashboard analytisc). I am NOT impressed. But I do think more solid education with the see stanford site will bridge some gaps there.

doing ML is now easier than ever. I worked on practical problems using ML back in the days when it was way less hyped than today. Back then we had to implement lots of "guts" of the process by ourselves. Now, you call caret or scikit learn or h20 or spark mllib with just a few lines, and "it" does everything for you from CV to pretty plots.
That said, by abstracting away a lot of the guts, i feel one step farther away from "how things work" and if i ever need to solve a problem that deviates from the default cases, it would require more thinking.

as a learning practice, "hand implement" some of the guts can help bridge this gap.

. ----starting with a good overall "view" definitely helps.

Kaggle is not everything, consider kaggle as a dictionary or playbook. Not all the gory details of a production system ML solution appears in Kaggle (lots of them, dont), and not all the complex tricks used in Kaggle warrants the effort in practice. Nonetheless, it's good to know how much headroom you have, and make a conscious tradeoff between ease/explainability and predictive power.

it's shocking how many people (even people in industry) seem to have no idea that not all DS problems at prediction problems. Look up the word "inference". Look up "planning".

灰常有帮助

quite helpful

添加一些关于OR的内容：
我自己也是OR盲，这部分仅供参考
https://www.quora.com/Are-there-good-online-courses-for-Operations-Research

Nikos Makrymanolakis, M.Sc., ph.d (cand.) in the area
Some very good and relevant courses about OR subjects in coursera:

* Discrete Optimization by Professor Pascal Van Hentenryck
* Algorithms, Part I by Kevin Wayne and Robert Sedgewick
* Algorithms, Part II by Kevin Wayne and Robert Sedgewick
* Algorithms on Graphs and Trees by Alexander S. Kulikov and Michael Levin
* Algorithms: Design and Analysis by Tim Roughgarden

Most of the algorithms covered in the above section, are OR used algorithms. The discrete optimization course is excellent, focus entirely on optimization (you will love the professor).

Feng Mai, ‎Assistant Professor at Stevens Institute of Technology

Operations Research is a broad field. For optimization I would recommend

Prof. Stephen Boyd's convex optimization (available on YouTube) and
Prof. Pascal Van Hentenryck's discrete optimization (coursera).

https://orc.mit.edu/academics/course-offerings

Somewhat older list
http://www.orcomplete.com/internet/enesbilgin/open-courses-on-operations-research

The list from stanford
https://see.stanford.edu/Course
.