someone asked me to make a test of R for the one applying a position in my previous company.
the question is something like:
1, what is a functional?
2, why vapply() is better than sapply()?
3, what is a function factory?
4, what is the difference between the return value of df[ ,"col"] if df is a tibble instead of data frame
5, what is S3 OOP in R? what is the difference between this and other OOP in other languages in Python?
Data Science之你問我地答
用戶1
680 回覆
30 Like
4 Dislike
其實呢度有幾多人係讀緊書
有幾多人係做緊呢行
有幾多人係做緊呢行
Look up Revnet.
In short, when computing gradients you have to recompute the output of the previous layer's output instead of storing them.
p.s. I had to such the same problem for a biomedical imaging company.
In short, when computing gradients you have to recompute the output of the previous layer's output instead of storing them.
p.s. I had to such the same problem for a biomedical imaging company.
用好多stats, computer science
this is the best question ever asked in this post 
for most people they can just get something done but they do not understand why it failed or succeeded, or at best vaguely understand certain sorts of methods available for particular scenario
for most people they can just get something done but they do not understand why it failed or succeeded, or at best vaguely understand certain sorts of methods available for particular scenario最易上手係邊個software?
Ching 點睇 poly data science 
我當你係指customer analytic既DA BA la
如果你話單純BA/DA既話我覺得都仲有得玩
career path我就唔太清楚, 至少我公司都仲未話有DS就唔要BA/DA
始終好多方法都行左幾十年,無可能話甩就甩
BA/DA好多時drill D細既data set drill得好deep
DS通常應該無咩興趣做番D傳統rule base野
同埋DS我覺得弱既係好多時都好依靠model
尤其係D 超級tech底既都鐘意砌好多複雜model
而呢D model 一般可能都要好多data量support
我諗大部份公司頂盡做customer analytic都唔會有多過50個features
係目前既情況我估3-5年內DA香港都仲係主流
如果你話單純BA/DA既話我覺得都仲有得玩
career path我就唔太清楚, 至少我公司都仲未話有DS就唔要BA/DA
始終好多方法都行左幾十年,無可能話甩就甩
BA/DA好多時drill D細既data set drill得好deep
DS通常應該無咩興趣做番D傳統rule base野
同埋DS我覺得弱既係好多時都好依靠model
尤其係D 超級tech底既都鐘意砌好多複雜model
而呢D model 一般可能都要好多data量support
我諗大部份公司頂盡做customer analytic都唔會有多過50個features
係目前既情況我估3-5年內DA香港都仲係主流
間間U黎黎去去都係教果D野
so far UST係香港第一個既Big data Master
本身有少少保證之外
其他果D (包括HKU) 我都覺得無咩大分別
睇你咩年紀咩財力
本身錢多年紀又後生既可以試下讀
如果好似小弟岩岩grad果時人工又低,又無積蓄我就唔會諗
講真DS degree真係一街都係
我唔覺得你比舊錢真係帶到比你咩實際價值
如果你有少少工作經驗相關既我諗就更加唔緊諗
我相信個data science degree唔係necessary, 純粹係過HR用
通常computer/stat degree都過到HR
so far UST係香港第一個既Big data Master
本身有少少保證之外
其他果D (包括HKU) 我都覺得無咩大分別
睇你咩年紀咩財力
本身錢多年紀又後生既可以試下讀
如果好似小弟岩岩grad果時人工又低,又無積蓄我就唔會諗
講真DS degree真係一街都係
我唔覺得你比舊錢真係帶到比你咩實際價值
如果你有少少工作經驗相關既我諗就更加唔緊諗
我相信個data science degree唔係necessary, 純粹係過HR用
通常computer/stat degree都過到HR
stochastic dynamic programming, optical control theory, convex optimization
容我潑一潑冷水Operations research做optimization 先最有用
我覺得好多人都想入行，但事實係真係幾難入，早兩年我入果時呢行未咁興，我都係誤打誤撞入左行，入左去兩個月先有個識做既人帶我地
今時今日求職者會遇到以下幾個case
1.無related master既人好易比hr screen走，你地都明hr唔識野，因為呢幾年好多呢d programme，黎緊幾年應該仲難爭
2. well established既公司entry位仲係好少
我有問我公司既senior點解唔請entry
佢話因為依家supply太多，而且係呢兩年真係多左呢d post 有實戰經驗既人都多。
3.搵工要小心
Title 叫data scientist，但做既野唔係
特別係份工提到要識hadoop/mapreduce/java，好大機會係起db/data engineer
ds主要係business上，好少咁infra，如果呢樣都要做埋唔係一個人可以做得哂既野
4. In你果個可能唔知自己想要咩人
某big 4 analytic team有次去的公司pitch我地，但係只係sell我地linear regression
之後有緣in佢地，佢interview問好多好OOP既野（咩係class，python點處理多程inheritance問題etc），多少少data strcuture。知左一定有幫助但做落真係幾用到，佢最後應該係好易請一班p仔，但唔識apply去business既人。
5.退而求其次，搵BA/DA做住先
唔係唔好，但我自己覺得skill set好唔周，同埋目前除左真係超大既公司，呢d role無條clear path，做完DA/BI係唔係真係轉到DS?唔一定，不過起碼係個起始點，而又真係有公司岩岩轉型之初會拉班DA試做POC
我唔係想放負，但大家都明真係好現實
而我覺得最難過都係HR果關
how many data scientists understand "deep" theory of the model they developed? how many of them are "regression monkey"?
我覺得其實好極端，多數有年資既DS都係識深入既theory，唔同models既特點, 因為依加相對成熟既DS team請人要求都唔低，會問得好深入或姐做tests。而亦有一堆所謂DS完全係掛名，甚至只用excel、唔識coding/stats都叫DS title