Data Science之你問我地答

DL 其實係by experiment 多過理論

贏得多唔難，要keep住贏先難

你唔知D野本身做咩點試
鳩試又真係OK既
花時間姐

感謝巴打實際既建議

亦有人分析過，武漢肺炎之後會有更多網上業務發展，所以data science既空缺會更多

你點睇呢樣野？

留名學習先

依家新仔0經驗入行咩價位
5年經驗左右又咩價位

你講得岩, 長嬴真係好難

hyper param 遠比network structure 重要

你唔係勁撚過提出yolo resnet vgg果班友掛

有個random connected network都係by experiment 試到準為止

實際計數仲會limited by hardware, 你有幾多gpu ram 俾你任用

其實data science 係咪一個buzzword 黎
個人覺得Machine learning 都已經buzz buzz 地，但都有好多正經既research ，data science 好似無乜

data science = applied statistics = applied applied mathematics

一係做statistician, data engineer, machine learning engineer, research scientist，呢D係專業

DS兩頭唔到岸，又要cs，又要stat，又要business knowledge ，又要reporting 緊係似風水佬

如果鍾意social，present 加少少tech 既人應該ok

Python係咪必需？
我社科出身，依家喺大學用R做data mining，諗住自學惡補數底同讀個data science master洗底入行

你識用d lib既c++ matlab都得, 問題係人地中學到大學果d linear algebra底都打到好穩, 計DL仲要識tensor operation, 你有無信心一年半年就跟得上進度先

半年真係只夠coding 入門咋

應該會有啲工唔洗既
但大路黎講Python係走唔甩，R越黎越少人黎用

個人見解，唔係真理黎

我都覺得網上行銷一定ML最大既business應水，始終係B2C，數據量方面in born一定多

不過個問題係，如果依家先開始搞既，中大型公司d data好多係唔同system上，點樣integrate埋一齊做好個ETL pipeline好緊要，infra當然野可以用cloud，但真係落手睇data始終都要人肉

所以我覺得黎緊好短期內，data engineer就一定好大需求。有公司搵現有人做，有公司外面請人，尤其係唔想cloud果d公司要搵人番黎搞infra。

2/3年後我相信都仲有條水可以食
依家入行如果你做data engineer識得執番靚d數，絕對係有優勢
但你最擔心既係當你scale大左，公司會搵條友番黎做model，你會同佢一齊砌model，但因為你backend skill 勁過佢，所以執DB砌ETL pipeline都主要係你做，佢就100%時間砌model

外資P

好多人方向都錯左
我都曾經係

若果有巴打有緣見到，希望我比到大家一點啟示。

data science/machine learning只係工具，所以，一係你就係製造工具既scientist，一係你就係使用工具既人.

如果你興趣係發明新工具，恭喜你，你應該好勁但係你要記住，你勁唔勁得過一個Lab？勁唔勁得過Google？有無AWS咁多business data？ 所以，認真做science既人會去讀PhD.

第二種係data science. 我要重申再重申，domain knowledge 係基礎，data science係工具。或者返轉來講，如果你係絕大多數行業（金融/地產/tech/marketing/engine/research/you name it），如果你唔開始學，五年之後你再問人咩係data science，D人會答你係 common sense 睇數據。

第三種係我希望大家學識既，係明白同利用年輕人既優勢。

我比個scienario大家，係我以前公司（ budge bracket）請ML人既code interview問題：

問：公司有好多web site, global business所以需要翻譯，你任務係design一個比Google或其他laboratory更accurate既NLP transformation model.

有興趣知我可以繼續😂唔係就當我9up啦

totally agree with ds is a kind of tool...

(各位sorry, 我打中文太慢, 英文會快好多, 所以出英文)

First of all, what is science in layman term?

data science can be divided into two area:
'science about data' & 'how data apply in business function'

in 'science about data', it include stat, modelling, systematize normality into a theory, invent new scientific method, etc. which heavily involved in academic area

'how 'data' can be applied in business function', like the people talk about in the street, how to make use of data, fit in to current business model, etc. focus in application.

many of us discussed a lot on which programming language we should learnt in order to get a entry of this industry... my opinion is any one is ok but you have to develop your sense of 'data' with it. I always talk with my mentee that programming language is just a 外功, the most valuable thing is how you solve the problem with your tool and skill which that problem solving skill is your 內功心法.

back to the topic, 'science about data' & 'how data apply in business function' is two area but they are interrelated with each other. Like 'DeepLearn' said, 'if you interested in building tool, go 'scientist' way; otherwise go 'analyst' way.

利申: degree in business stat, worked in an insurance company 10 years ago, now in junk bay as a data modeller/engineer

我本身有C++同machine learning 底，因為大學讀過，但係好廢同好少玩prgramming
依家開始的起心肝想學programming for machine learning。應該由邊樣野開始入手。
我用過google big query, azure, aws, Neural desingers, matlab, IBM既machine learning toolkit, 姐係唔洗打code 咁濟。
近排試下由Jupyter notebook 入手，因為好方便，但係佢比我揀python pytorch 我已經唔知自己揀乜好
之後自己裝左anaconda, 我裝左好多野，個Roots 有 base, python-gpu, tensorflow- gpu 又搞到一頭冒水，巴打指點下我應該之後落黎係點去學好。。。。

有無做面試考官?
如果可唔可以分享吓D深刻故仔