【在线Info Session】业界数据科学家回答你关于找数科工作的问题

小K 发表于 2015-5-24 03:26
If you don't have an intern offer now, it is slim chance you will find an internship this summer a ...

Thank you so much. I like these straightforward answer and suggestions.

For count variable prediction, intuitively I will try possion regression first, but if there is overdispersion (sd>>mean), we need to use NB regression.
For RF, basically it could improve acuracy of prediction and provide almost unbiased estimation. But also RF might lead to overfitting, and the model might not be interpreted.

Yes, I will move to more data structure, algorithm and leetcode things. Also spark things.
For your following question, I'd like to say, yes, I really enjoy data science work, especially coding and finding ways to improve acuracy, or digging more interesting things from data.

Could I ask two more questions?
1. for companies in IT (most familiar business for me), what's the most important language/tool for data science? R or Python, or other? Because I'd like to focus on one.
2. do you think data visualization skill is important for data science? since you didn't mention it in your background survey questions. I was told data visualization skill is very important, especially for novice.

eve2ada 发表于 2015-5-24 08:11
Thank you so much. I like these straightforward answer and suggestions.

For count variable pred ...

not sure I'd agree with the RF part :) You can read more about this. And ML in general.

1. DS is not a one tool trade --- NOT AT ALL.
If you think you can learn one and be all set....well you will be very disappointed.

again I shamelessly refer to my own post:
http://www.1point3acres.com/bbs/thread-76429-1-1.html

A PhD Data Scientist: Jack of All trades, master of one.

您好！
本帖隐藏的内容需要积分高于 133 才可浏览
您当前积分为 0。
使用VIP即刻解锁阅读权限或查看其他获取积分的方式

游客，您好！
本帖隐藏的内容需要积分高于 133 才可浏览
您当前积分为 0。
VIP即刻解锁阅读权限 或 查看其他获取积分的方式
Unlock interview details and practice with AI
Curated Interview Questions from Top Companies

No in the sense that you don't have to master highcharts (js), d3 to work as DS. It certainly would be fun and helpful. Think new york times and flowingdata.
..
For dashboard building, nowadays there are many ready made tools to help. for analyst type or data engineer type, they'd use these tools a lot.

So viz for exploration and story telling, yes. viz for dashboard is not that hard, viz for production quality stuff facing outside customers? You can often get help from front end engineers.

小K 发表于 2015-5-24 11:50
not sure I'd agree with the RF part :) You can read more about this. And ML in general.

1. DS  ...

Thanks a lot for your time.
. ----
I am not PhD actually, and I ever thought of it. But because of my industry background I found myself more prefer to learn in real business world instead of academic environment.
I agree with you on "Jack of all trades, master of one", and I am trying to cover "all trades" from the beginning I entered DS domain. Now I think I am locating "the one" :) I assume it will be python or java considering the business environment despite that I love R....

And yes,  I will learn more on randome forest.

Thanks again for your precious weekend morning:)

小K 发表于 2015-5-24 11:50
not sure I'd agree with the RF part :) You can read more about this. And ML in general.

1. DS  ...

And sorry for an additional question:
about data structure and algorithm, could you describe roughly that what would be most important or covered in data science? or just some examples?

Thanks again.

自我介绍：Fisrt year PhD Candidate@CS,北美某小学校，CS氛围不浓厚, 导师大数据安全方向（刚起步，研究方向不明朗）
背景 统计信号处理，国内基本是做信号检测与估计的，熟悉Markov Chain Monte Carlo, Regression，熟悉大部分基本的统计知识. 个人感觉ML与统计信号处理非常接近，可以做一般算法的公式推导，看懂一般理论文章压力不大。
代码方面会什么，到什么程度: Python学习中，感觉很强大，熟悉Matlab. R还没有涉足。Javascript，java，C++略懂。已经修过算法课，并且当过本科生Algorithms的TA，马上开始Leetcode刷题。
你知道你在面试之前不可能搞定的是：Hadoop等分布式计算技术。主要是目前缺乏平台学习。
希望成为既懂得系统架构又懂得数据分析的全栈Data Scientist.

xuhao417347761 发表于 2015-5-24 13:48
自我介绍：Fisrt year PhD Candidate@CS,北美某小学校，CS氛围不浓厚, 导师大数据安全方向（刚起步，研究方 ...

Sounds like a very well rounded education.
Hadoop可以自己设置一个本地的试试手。不难。udacity的相关课可以看看。不是很费时间。
可能多动手做项目就有用吧，你的背景挺好的。
统计好的CS人会有非常强大的发展。
good luck!

谢谢小K

==================
你的专业，学位，毕业时间？
本科Electrical Engineering, 作为software engineer在国内工作过两年，研究生Engineering Management, 项目比较business，我选课都尽量往data方向靠，2015 fall毕业

你在统计方向对哪些问题很有把握，哪些不太有把握？
有把握的：hypothesis testing，glm，regression, bayesian
不太有把握的： time series, 以及模型背后的assumption, inference, forecasting

你在代码方面会什么，到什么程度（数据结构，算法，是否刷过题）？
数据结构基本都还记得，但是需要一个星期来refresh一下
算法已经很生疏了，还没刷过，打算下学期强化一下


你在机器学习方面，除了调用现成包之外，还知道哪些？
基本都是自学的，从Andrew Ng的Machine Learning，到stanford的Mining Massive Dataset， 再到各种blog，ipython notebook
对常见的regression和classification的特点，局限，使用场景比较熟悉，也跟着notebook做过一些小project

你最有把握的语言、工具是：R, python

你可能熟悉的语言、工具是：C, Java

你知道你在面试之前不可能搞定的是：不可能成为熟练的码农，了解设计模式，软件工程

你觉得你的business sense如何，这里的几个问题你有什么想法吗？
我觉得跟科班stats比起来我的长处在business sense. 之前一次比较大型的3校联合data analysis competition我们胜过那些专业出身的获得冠军，我觉得原因就是我们的模型虽简单，但Make more sense to sponsor's business

你希望找什么样的工作.
暑期在一家consulting firm做data scientist intern. 但以后还是想去technology firm, 比如facebook，linkedin这样的地方做DS
我觉得自己的劣势可能在于1. 专业背景（非stats非phd） 2. 对于technology firm，代码能力还不够

非常感谢！十分期待听到你的意见。

k妈能帮忙改简历吗

你的专业，学位，毕业时间CS/STAT MS 2016

你在统计方向对哪些问题很有把握，哪些不太有把握
只上过一些基本的统计，没有深入，也上的不太系统。没时间也没精力去好好系统的上统计理论课，只选了一些实用的课上。喜欢algorithom based的统计，用代码去实现。


你在代码方面会什么，到什么程度（数据结构，算法，是否刷过题）
上过算法，准备看下数据结构，还没刷过题


你在机器学习方面，除了调用现成包之外，还知道哪些
ML水很深，感觉MS学的基本都是很皮毛的东西，也就不在这班门弄斧了

你最有把握的语言、工具是：
R/Python

你可能熟悉的语言、工具是：
Java/Javascrit/PHP

你觉得你的business sense如何，这里的几个问题你有什么想法吗
除了本科上过1-2门课没啥其他背景，不过对这个方向还是有兴趣，也想尝试下

你希望找什么样的工作:
可能跟business analytics类似的工作

小K 发表于 2015-5-24 16:21
Your background seems more coding oriented than stats, so work with that. Many companies do not re ...

谢谢小K姐呀：

选择数科的理由：
1、兴趣，一直享受着用理论知识，从大数据中寻找解决实际问题的方法，编程实现并创造价值的乐趣。
2、机遇，相信数科领域的前景很广阔

学好机器学习的方法：有志同道合的伙伴一起探讨，共同学习，并且能够迅速在项目中实战检验。具体做过的项目有推荐系统，文本分类和反作弊用户识别。

===========================

由于入门不久，简历还很捉急，不知道要如何润色体现竞争力，小K姐求指点