查看完整版本 : Data Science Master讀唔讀得過?

anson2004 2020-1-2 01:20 PM

Data Science Master讀唔讀得過?

見本地十幾至廿幾萬.


但呢科上網好多resource,
coursera , edx , udacity, Udemy,kaggle, kdnuggets, datacamp 都有,有D甚至免費


又有CHING講過呢科其實只係ug stat level, 再加用tools .
聽講taught master都不過係convex optimization, bayesian stat, neural network.
呢D UG maths/stats level都有教的野.
感覺呢科似bootcamp(例如visualization,data cleaning,data mining, wrangling)多過似獨立學問
除左張沙紙,呢科值唔值得讀?

[[i] 本帖最後由 anson2004 於 2020-1-2 01:31 PM 編輯 [/i]]

narius 2020-1-2 02:15 PM

Obviously depends on whether you are getting good instructors. Master level classes are really not about the core ideas, but training students to apply.


Take something very simple like linear regression. It turns out most students (well, at least mine .. and I am talking about master levels) have little clue about how to use it well. 


It is not about the basic technique (if you use a tool like R or Python, it is a trivial 5 min exercise) but about how to put the model together, iterate based on preliminary analysis, and judge what the next step approach to be. I find even PhD students sometimes have problems with using stats and data science effectively.


Knowing a technique is trivial (google has all the materials). Internalizing the intuition and effective application is a bit harder. The hardest is to re-combine and/or formulate new approaches. For example, linear regression is often used in multiple stages. Understanding how and why (why is more important than how) in each situation is a bit more involved than your undergraduate 101 classes.


And that is *just* regressions. There are a thousand things beyond that. 

[[i] 本帖最後由 narius 於 2020-1-2 02:16 PM 編輯 [/i]]

anson2004 2020-1-2 09:14 PM

[quote]原帖由 [i]narius[/i] 於 2020-1-2 02:15 PM 發表 [url=https://www.discuss.com.hk/redirect.php?goto=findpost&pid=512533275&ptid=28793744][img]https://www.discuss.com.hk/images/common/back.gif[/img][/url]
Obviously depends on whether you are getting good instructors. Master level classes are really not about the core ideas, but training students to apply.


Take something very simple like linear reg ... [/quote]
我聽過師兄大名,你係大學做教學同研究,
一定係呢方面好有熟悉, 以上你的提點多謝指教:D
我岩岩完成左coursera ML ,
睇完stanford CS229 D PAPERS,
都覺DATA SCIENCE好多野學,
其實theoretical DS好有趣
岩岩睇完PCA同Eigenvector關係已經令我大開眼界

純粹想比較一個MSc.
同市面上edx, udemy, coursera 呢d雞精courses
加自己買書睇e.g. Introduction to statistical learning, boyd's convex optimization .Morgam kauffman Data Mining,結果是否會一樣:o

[[i] 本帖最後由 anson2004 於 2020-1-2 09:22 PM 編輯 [/i]]

竹劍 2020-1-2 09:52 PM

[quote]原帖由 [i]anson2004[/i] 於 2020-1-2 01:20 PM 發表 [url=https://www.discuss.com.hk/redirect.php?goto=findpost&pid=512530982&ptid=28793744][img]https://www.discuss.com.hk/images/common/back.gif[/img][/url]
見本地十幾至廿幾萬.

但呢科上網好多resource,
coursera , edx , udacity, Udemy,kaggle, kdnuggets, datacamp 都有,有D甚至免費

又有CHING講過呢科其實只係ug stat level, 再加用tools .
聽講taught master都不過係convex optimization, bayesian stat, neural network.
呢D UG maths/stats level都有教的野.
感覺呢科似bootcamp(例如visualization,data cleaning,data mining, wrangling)多過似獨立學問
除左張沙紙,呢科值唔值得讀? [/quote]
睇你邊間邊個department讀(學校出唔出名, 教既內容教得深唔深入)
仲有睇你本身讀/過識D乜....

narius 2020-1-3 01:46 AM

[quote]原帖由 [i]anson2004[/i] 於 2020-1-2 09:14 PM 發表 [url=https://www.discuss.com.hk/redirect.php?goto=findpost&pid=512549362&ptid=28793744][img]https://www.discuss.com.hk/images/common/back.gif[/img][/url]

純粹想比較一個MSc.
同市面上edx, udemy, coursera 呢d雞精courses
加自己買書睇e.g. Introduction to statistical learning, boyd's convex optimization .Morgam kauffman Data Mining,結果是否會一樣 [/quote]

Good question. Let me use the philosophy of a recent executive data science class (executive meaning that we tailored a program for a local large company) to answer your question. The first thing i told the students is that i do NOT teach out of a textbook. The simple reason is that if you just want the information, google or just reading a book is as good as me.

So what is my value add? The answer is in how i run the class. The class is 100% project based. I am a firm believer that one can truly learn (meaning internalize all the necessary nuance and intuition) via doing. Hence, i often say, i do not teach .. but i train students. Now, to be fair, in that course i did have to give some lectures since there are certainly knowledge gaps in the students. But what I tried, aside from teaching the most basic, is to give the idea of a technique and as importantly .. what to search for if you need the details. And in a majority of the time, i basically act like a consultant and coach on the students' own project. So the value add is the discussions, the Q&A, and the coaching ... not very different from the consulting work i am doing with silicon valley companies, or how I work with PhD students.


I trained my son the same way, when he was in high school. He got to learn a few variations of regression methods by doing a survey project on Amazon mechanical turk.

Take PCA as an example. I am glad that you appreciate how it works. But the point is really not the specifics of the method but the larger idea of thinking in a high dimensional space .. and more importantly, the idea of dimensional reduction. All you are doing is to find a subspace (which the eignevectors spans) that holds a large fraction of the relevant information.
If you think that way, a lot of other methods become clear ... even simpler things like variable selection. BTW, *any* stat and data science methods can be viewed as geometry in high (sometimes infinite) dimensional space. It is much more important to internalize these concepts and view all the methods as a whole. No one can remember and know all the possible methods out there. When we do *real* work, often we pick up new methods on the fly so the ability to understand how something works by reading about it in 15 min, is almost a requirement to be successful in the data science (or any scientific) business.

So my advice is this. It is certainly great to learn via different focus courses and youtube ... but find a project to do. It can be anything from a data science challenge to a personal project. Data is everywhere you look. Secondly, go for a master program only if there is a hefty emphasis on projects and personal coaching.

Now, what i say is in the perspective of learning. Sometimes you do need the right qualification to break into the industry. While having done a project you can discuss and be able to solve problems go a long way, a degree may be needed to make sure that you are not filtered out in the first round.

竹劍 2020-1-3 03:10 AM

[quote]原帖由 [i]anson2004[/i] 於 2020-1-2 09:14 PM 發表 [url=https://www.discuss.com.hk/redirect.php?goto=findpost&pid=512549362&ptid=28793744][img]https://www.discuss.com.hk/images/common/back.gif[/img][/url]
純粹想比較一個MSc.
同市面上edx, udemy, coursera 呢d雞精courses
加自己買書睇e.g. Introduction to statistical learning, boyd's convex optimization .Morgam kauffman Data Mining,結果是否會一樣:o[/quote]
學得好唔好先唔講. 問題number one: 你有冇需要要有一個data science or etc之類既degree放係CV上面搵工?

anson2004 2020-1-3 09:28 AM

[quote]原帖由 [i]narius[/i] 於 2020-1-3 01:46 AM 發表 [url=https://www.discuss.com.hk/redirect.php?goto=findpost&pid=512556864&ptid=28793744][img]https://www.discuss.com.hk/images/common/back.gif[/img][/url]


Good question. Let me use the philosophy of a recent executive data science class (executive meaning that we tailored a program for a local large company) to answer your question. The first thing  ... [/quote]
Thank you so much!, you'd provided much insight to me
You are a star !

[[i] 本帖最後由 anson2004 於 2020-1-3 09:29 AM 編輯 [/i]]

anson2004 2020-1-3 09:31 AM

[quote]原帖由 [i]竹劍[/i] 於 2020-1-3 03:10 AM 發表 [url=https://www.discuss.com.hk/redirect.php?goto=findpost&pid=512557694&ptid=28793744][img]https://www.discuss.com.hk/images/common/back.gif[/img][/url]

學得好唔好先唔講. 問題number one: 你有冇需要要有一個data science or etc之類既degree放係CV上面搵工? [/quote]
都岩,呢個係去到好現實問題, Msc.呢個"功能"對入行搵工的幫助的確要考慮埋

IWC-Rolex 2020-1-3 10:58 AM

讀得Taught Master,就一定要研究一下學校就業報告,睇睇以前用自己差不多背境既學兄,學姐,畢業後去咗邊,雖然話最後都係睇人,但呢個都係唔錯指標。

竹劍 2020-1-3 06:07 PM

[quote]原帖由 [i]anson2004[/i] 於 2020-1-3 09:31 AM 發表 [url=https://www.discuss.com.hk/redirect.php?goto=findpost&pid=512563769&ptid=28793744][img]https://www.discuss.com.hk/images/common/back.gif[/img][/url]
都岩,呢個係去到好現實問題, Msc.呢個"功能"對入行搵工的幫助的確要考慮埋 [/quote]
除非你已經有個好相關既學位, e.g. MSc in CS / Stat 之類, 否則master其實唔太走得甩.

你去coursera , edx , udacity果類上唔係唔得, 但係其實齋睇video/書唔夠, 要落手做埋個project/功課先叫完成左個basic, i.e. 一樣要交錢.
條數計埋計埋, 上萬唔係夢. 同樣花咁多時間, 不如畀多少少整埋個學位好過.

By the way, 本地邊件Data science / data whatever既 master十幾萬埋到單? 唔係冇, 我知邊兩件係, 有一件ok但係偏mathematics, 另一件就建議三思...

至於教得有幾深入, 就睇課程點設計了. 同樣叫data science之類, 唔同學校可以相差好遠的.

janeleewc 2020-1-3 07:20 PM

[quote]原帖由 [i]anson2004[/i] 於 2020-1-2 09:14 PM 發表 [url=https://www.discuss.com.hk/redirect.php?goto=findpost&pid=512549362&ptid=28793744][img]https://www.discuss.com.hk/images/common/back.gif[/img][/url]

我聽過師兄大名,你係大學做教學同研究,
一定係呢方面好有熟悉, 以上你的提點多謝指教:D
我岩岩完成左coursera ML ,
睇完stanford CS229 D PAPERS,
都覺DATA SCIENCE好多野學,
其實theoretical DS好有趣
岩岩睇完PCA同Eigenvector關係已經令我大開眼界

純粹想比較一個MSc.
同市面上edx, ud ... [/quote]

你諗住讀邊間 可唔可以pm

Vicson 2020-1-5 04:50 PM

想問問如果無programming底 但有d math同stat底 應唔應付到DS master...

icantstop 2020-1-5 09:40 PM

[quote]原帖由 [i]anson2004[/i] 於 2-1-2020 01:20 PM 發表 [url=https://www.discuss.com.hk/redirect.php?goto=findpost&pid=512530982&ptid=28793744][img]https://www.discuss.com.hk/images/common/back.gif[/img][/url]
見本地十幾至廿幾萬.


但呢科上網好多resource,
coursera , edx , udacity, Udemy,kaggle, kdnuggets, datacamp 都有,有D甚至免費


又有CHING講過呢科其實只係ug stat level, 再加用tools .
聽講taught master都不過係convex optimization, bayesi ... [/quote]

伏!!😵

narius 2020-1-6 04:10 AM

[quote]原帖由 [i]Vicson[/i] 於 2020-1-5 04:50 PM 發表 [url=https://www.discuss.com.hk/redirect.php?goto=findpost&pid=512663529&ptid=28793744][img]https://www.discuss.com.hk/images/common/back.gif[/img][/url]
想問問如果無programming底 但有d math同stat底 應唔應付到DS master... [/quote]

Clearly it depends on the person. From what i have seen, learning enough programming for doing data analysis is pretty trivial. I have yet seen a person with good technical background who cannot do it.

Most of my students can pick up a language (say R or Python) in a week or two. Having said that, i only teach an elective master level class which i limit to around 10, and train PhD students. And I have heard horror stories from colleagues who teach large programming classes (even at the master level!) that students cannot do simple programming things. But that my be motivation issues, not that any of it is difficult.

The easiest thing to do is to find a R or a Python tutorial, follow throw the sample problem and try a small project. It is not something that you need to ask strangers about. You can easily find out.

竹劍 2020-1-6 05:16 AM

[quote]原帖由 [i]Vicson[/i] 於 2020-1-5 04:50 PM 發表 [url=https://www.discuss.com.hk/redirect.php?goto=findpost&pid=512663529&ptid=28793744][img]https://www.discuss.com.hk/images/common/back.gif[/img][/url]
想問問如果無programming底 但有d math同stat底 應唔應付到DS master... [/quote]
冇programming底就去學.

而且, 你咁問法好令人懷疑係有幾多math/stat底.
正常黎講, math/stat major出黎既, 應該點都識少少programming, e.g. MATLAB or R咁的.

Vicson 2020-1-6 10:15 AM

[quote]原帖由 [i]narius[/i] 於 2020-1-6 04:10 AM 發表 [url=https://www.discuss.com.hk/redirect.php?goto=findpost&pid=512686290&ptid=28793744][img]https://www.discuss.com.hk/images/common/back.gif[/img][/url]


Clearly it depends on the person. From what i have seen, learning enough programming for doing data analysis is pretty trivial. I have yet seen a person with good technical background who cannot d ... [/quote]
我本身econ major,係有用過R去做教過regression,亦覺得應付到...但問題是唔知econ教programming的程度足唔足夠令我應付到master in DS的level.....始終compare to CS/Engineer/STAT major個d, ug econ教的programming程度我知一定係好入門的程度...

竹劍 2020-1-6 05:23 PM

[quote]原帖由 [i]Vicson[/i] 於 2020-1-6 10:15 AM 發表 [url=https://www.discuss.com.hk/redirect.php?goto=findpost&pid=512692396&ptid=28793744][img]https://www.discuss.com.hk/images/common/back.gif[/img][/url]

我本身econ major,係有用過R去做教過regression,亦覺得應付到...但問題是唔知econ教programming的程度足唔足夠令我應付到master in DS的level.....始終compare to CS/Engineer/STAT major個d, ug econ教的programming程度我知一定係好入門的程度... ... [/quote]
咁就好depends on 你邊個DS master了.

如果你R算用得順手既, 估計到左DS master用python果陣唔會係乜大問題.
但係如果佢玩到scala for big data, 或者interactive data visualization用javascript果類, 就可能煩多少少了.

[[i] 本帖最後由 竹劍 於 2020-1-6 10:13 PM 編輯 [/i]]

narius 2020-1-7 10:38 AM

[quote]原帖由 [i]Vicson[/i] 於 2020-1-6 10:15 AM 發表 [url=https://www.discuss.com.hk/redirect.php?goto=findpost&pid=512692396&ptid=28793744][img]https://www.discuss.com.hk/images/common/back.gif[/img][/url]

我本身econ major,係有用過R去做教過regression,亦覺得應付到...但問題是唔知econ教programming的程度足唔足夠令我應付到master in DS的level.....始終compare to CS/Engineer/STAT major個d, ug econ教的programming程度我知一定係好入門的程度... ... [/quote]

Do you just run vanilla regression or can you program a little more complicated methods? For example, can you write a program to do an instrumental variable two-stage regression with a general data set (yes, i know there are standard libraries to do IV, but i always ask my students to do it themselves so that they understand what is going on).

The other simple test is to see if you can duplicate your R code in python and how long you need to do that. I would recommend you use anaconda or the mini-conda version, and put Jupyter Lab (or Jupyter Notebook) as the front-end.
頁: [1]
查看完整版本: Data Science Master讀唔讀得過?