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

要好好学习

有人如此指点真是好幸运

sister little K~

能告诉我们您的背景嘛（比如本科学校，GPA），想和牛人对比一下，了解下差距。 论坛搜过了，貌似没相关信息耶~

walkingpup 发表于 2014-8-25 08:04
sister little K~

能告诉我们您的背景嘛（比如本科学校，GPA），想和牛人对比一下，了解下差距。 论坛搜 ...

本科学校一般。
GPA。。。。。。！！！！早不记得了。。。肯定不到90
但是我这贴跟我本科显然一分钱关系都没有啊。。。。

吃饭的时候看了一下udacity的software development life cycles这门课，actually非常不错，对于小白来说，填补了知识点上的一些漏洞。
免费课件可能缺少练习，但是介绍了什么是IDE（对啊。。。这种问题。。。。也得有地方讲啊。。。。），正规定义了一些平常听了无数次但是不完全清楚的词（比如scrum）。

背景：学过C, javascript，不过是本科前两年，基本忘光。MATLAB用的多，少量STATA的经历（不知道这软件自由度这么低学校为啥这么喜欢）。EE PHD+ECON MASTER DOUBLE MAJOR, some experience in econometrics
目标：一年之内学完Python, Java, R, HTML5, Javascript, CSS, Machine Learning, MapReduce, SQL
大方向：Python和Java预计花时间最多，现在开始学习熟练。R不打算花巨多时间，准备有个大概的了解。HTML5系列明年再处理，到时候准备租个网站边学边弄。Machine learning准备用Python来实现（主要看这个帖子：http://blog.renren.com/share/231 ... ose_time=1410188191），会花一定时间。剩下的暂时计划不到。
短期计划（3个月）：Python已学习1个月，上了google的课程+小K贴出的算法和数据结构（graph没学完），自己写code把数据结构都实现了一遍（CS同学说主要就学数据结构）。学了recursive programming后花了一天写了个解数独的程序。graph学完后转战<Python for data analysis>，同时开始上手JAVA(用Core Java的书),并用JAVA实现基础数据结构。零碎的时间看看以前C的课件，主要熟悉pointer，然后看一些介绍概念的公开课。
Note: 其实最开始的时候很抵触coding，不过现在无奈coding就像是会开车一样，不知不觉变成了一个必须的技能。data science需要，就连EE的硬件也希望你会，于是我也找不到继续逃避的理由了。SIGH

嗯，更新了一下首楼，主要是，现在觉得，”会一点编程的统计师“，就业出路估计没有“会一丁丁统计的码农”的1%，更新了技能stack
基本上，合格的码农和一些基本解决问题的能力，will take you  WAY, WAY farther than your statistics skills

虽然完全不懂统计是不行的，但是只懂一点点编程更是绝对不行的。

仅仅刷题是不够的，代码写的不够好，都会非常难过。
==============
吐血学编程ing

这么好的东西现在才看到，太可惜了，必须MARK!!!

i am no longer new to git, but i gotta say, the new udacity git course is BY FAR the best, beats all the rest of such tutorials !! and I love that two girls are teaching this! :)
Focusing on concepts is the way to go! most tutorials just teach you what commands to use without really explaining well why exactly you are typing these commands (with the exception of ProGit, but reading that book is not as easy as following this course!!)

also i'd like to add, the newer generation of courses really GOT it.
they are gradually filling the gap that I felt in my knowledge base, the gap that someone from a background other than SE/CS would find really hard to describe and hard to fill on one's own. Since most DEVs already know about it, they forgot it's something that has to be learned.

These will do LOTS of good, the earlier you learn them, the better. No need to be an expert in these areas, a general knowledge will carry far:

1. software engineering (udacity), even if you are not going to work in an IT industry. The stuff you learn from this kind of course will make you about way more productive, and save you lots of frustration.

1.5 Git and general workflow type of questions. This is not sth well taught in academia, especially if you are a PhD from non CS background. The Coursera JHU reproducible research part is a good start.

2. OOP (yes, I know Py does it too, but for sure, learn java), start from udacity or berkeley 61b.

3. Data science track of Udacity and Coursera - pick and choose, quickly go through. Key: know what's out there, and where to look for help

4. Web dev track of udacity: no I am not a web dev at all, but eventually a data scientist will work on data products. I don't think many companies hire statisticians to do pure research and NOT touch the "real stuff". Know the core competencies for sure, but also know how to present findings with interactive prototype and presentations. Most often the business need more than your keynote or ppt. Build prototypes. No, they are not only reporting dashboards. Even with your complex models, you need to show the results to decision makers. So know your html/css for sure, know your js, know your jquery, angularjs, jquery-ui, bootstrap etc. Yes know your data.table/ggplot/knitr, but also know your d3. Know them so you can quickly whip up something flashy and fancy. They like to tell you "looks" dont matter. They deny it but they are ALWAYS swayed by looks!!!

================
For BS/MS level dev, people like to emphasize algo, data structure etc. But I think for DS, the emphasis is less on being highly proficient in a small area. Instead you are supposed to be "fairly good" with a much wider area. Less restrictions and more freedom to try out different things. But also a lot more to learn. The problem space is more vague, less well defined, and don't yet contain a well known set of rules. The prep process, as opposed to, CC150+leetcode, may contain going through 5X as much material, but none too deep, EXCEPT for your core -- optimization for you AppliedMath/OR PHDs, stats for you STAT PHDs, ML for ML PHDs etc. Yes you do need a solid core....

Various EDU ventures are trying to change this. So excited to be part of this.




. Χ