Data Science之你問我地答
用戶1
680 回覆
30 Like
4 Dislike
冇乜邊個一開波就乜都識
跟住興趣學可以行到好遠
利申: 打code 路過
跟住興趣學可以行到好遠
利申: 打code 路過
乜類型都有可能嗰喎，睇嗰requirement 㗎喳，當一啲嘢唔知pattern,你cater 唔倒佢咁多種possibility嘅事候，但佢又有啲input/output example 喎，咁未好自然會用ML approach。呢類case 常見於啲develop 啲automation tool
For 你個case ，如果你又未學過networking（即係OSI 7層layer都唔知係乜）咁就要讀networking先啦，唔係嘅都可以take 吓唔同topic on machine learning 嘅，其實我覺得睇你覺得邊樣可以自己pick up 倒，邊啲要take course 學，turn out 要efficient 同effective 咁做嘢，你都係要識晒
我都係undergrad 讀CS 嘢，做咗幾年developer 先再讀master (嗰master 都唔係DS, 只係裡面有1/3 係 ML, 其他係 classical computer vision同control theory)
其實唔洗諗到做DS工真係要識好多嘢先，因為真係難嘅，佢都請PhD 做啦。hard skill 真係一兩個course 就學晒，再學就睇topic㗎啦，image? NLP呢啲睇自己興趣，同埋我覺得嗰重點係學到有能力睇paper, 因為呢個field update 得太快，你花時時睇paper，或者睇啲雞精blog post 畀你take course 嘅嘢仲多仲齊仲update
玩Kaggle 與其話係build profolio ,
不如當佢係用黎proof 自己真係apply 倒學過嘅嘢就算啦，你唔係高rank 其實唔太care
For 你個case ，如果你又未學過networking（即係OSI 7層layer都唔知係乜）咁就要讀networking先啦，唔係嘅都可以take 吓唔同topic on machine learning 嘅，其實我覺得睇你覺得邊樣可以自己pick up 倒，邊啲要take course 學，turn out 要efficient 同effective 咁做嘢，你都係要識晒
我都係undergrad 讀CS 嘢，做咗幾年developer 先再讀master (嗰master 都唔係DS, 只係裡面有1/3 係 ML, 其他係 classical computer vision同control theory)
其實唔洗諗到做DS工真係要識好多嘢先，因為真係難嘅，佢都請PhD 做啦。hard skill 真係一兩個course 就學晒，再學就睇topic㗎啦，image? NLP呢啲睇自己興趣，同埋我覺得嗰重點係學到有能力睇paper, 因為呢個field update 得太快，你花時時睇paper，或者睇啲雞精blog post 畀你take course 嘅嘢仲多仲齊仲update
玩Kaggle 與其話係build profolio ,
不如當佢係用黎proof 自己真係apply 倒學過嘅嘢就算啦，你唔係高rank 其實唔太care
啱呀，大學教育係畀你有自學嘅能力，之後都係睇自己。
如果你覺你將來做DS份工係唔洗起data pipeline, 咁都可以唔識networking 嘅
。雖然好多嘢而家都有現成software幫你做，但你點都要識啲基本嘢先睇得明啲doc, 出事識troubleshoot, 而唔係坐咗係度
modeling 其實係成個machine learning project 佔嘅部份得3成左右，其他都係IT嘢。
玩Kaggle 就係幫你練嗰3成，其他係學唔倒㗎。
當然你好肯定你將來份工只會handle modelling ( 再加埋啲畫BI, 同business user 解釋），咁係可以唔理除咗programming 以外嘅IT嘢嘅
但係唔係可以𠝹到咁開，嗰trend 係唔係𠝹開，同埋邊類人叫價能力高啲，就要你自己思考吓啦。
我自己就覺得，如果唔想咁多IT 成份嘅，你只可以揾啲已有完善data pipeline 嘅公司，你仲要有business sense/domain knowledge, 用data 幫間公司make decision, 甚至係幫間公司諗有乜問題可以用data 去揾答案，但呢啲business sense/domain knowledge 唔係讀書就得，而呢啲亦係所謂嘅吹水嘢，但係唔係真係人人都吹得起呢
另外，就香港黎講，好多好似好大嘅公司，其實極基量都係呢一兩年先開始build data warehouse, 仲要係冇乜insight, 人有我有嘅心態去起,
更加唔講佢啲internal process 好多仲用緊scan image 入pdf 就當digitize 咗，完全唔係machine-readable，你就會覺得香港搞DS, 其實要IT人多過要DS人
如果你覺你將來做DS份工係唔洗起data pipeline, 咁都可以唔識networking 嘅
。雖然好多嘢而家都有現成software幫你做，但你點都要識啲基本嘢先睇得明啲doc, 出事識troubleshoot, 而唔係坐咗係度modeling 其實係成個machine learning project 佔嘅部份得3成左右，其他都係IT嘢。
玩Kaggle 就係幫你練嗰3成，其他係學唔倒㗎。
當然你好肯定你將來份工只會handle modelling ( 再加埋啲畫BI, 同business user 解釋），咁係可以唔理除咗programming 以外嘅IT嘢嘅
但係唔係可以𠝹到咁開，嗰trend 係唔係𠝹開，同埋邊類人叫價能力高啲，就要你自己思考吓啦。
我自己就覺得，如果唔想咁多IT 成份嘅，你只可以揾啲已有完善data pipeline 嘅公司，你仲要有business sense/domain knowledge, 用data 幫間公司make decision, 甚至係幫間公司諗有乜問題可以用data 去揾答案，但呢啲business sense/domain knowledge 唔係讀書就得，而呢啲亦係所謂嘅吹水嘢，但係唔係真係人人都吹得起呢
另外，就香港黎講，好多好似好大嘅公司，其實極基量都係呢一兩年先開始build data warehouse, 仲要係冇乜insight, 人有我有嘅心態去起,
更加唔講佢啲internal process 好多仲用緊scan image 入pdf 就當digitize 咗，完全唔係machine-readable，你就會覺得香港搞DS, 其實要IT人多過要DS人
覺得而家網上所有online courses都主打ML，DS，AI，其實真實商業世界有幾大需求？
我諗99%既公司根本冇data可言，冇data又點需要請呢方面既人才？
讀呢d野好多人都係跟風。
我諗99%既公司根本冇data可言，冇data又點需要請呢方面既人才？
讀呢d野好多人都係跟風。
3成已經算多
睇過篇paper係咁形容ML system
睇過篇paper係咁形容ML system
fresh grad可以搵咩公司 搵緊工 冇頭緒
之前project做過賭馬 但太多數據都冇 所以睇得唔多 用SAS eg做既
睇咩project啦，雖然我講完三成左右都做度有人post 呢張圖㗎啦 ，想留返啲幻想空間...
，想留返啲幻想空間...我睇法同你相反
做得DS/AI/ML一係就大公司，一係就startup
大公司一定好多data, 唔係既話啲data infra公司唔會賣得咁貴，賺咁多錢, 例如Oracle
至於startup就自然會諗計搵data，用public data set, 自己crawl等
做得DS/AI/ML一係就大公司，一係就startup
大公司一定好多data, 唔係既話啲data infra公司唔會賣得咁貴，賺咁多錢, 例如Oracle
至於startup就自然會諗計搵data，用public data set, 自己crawl等
藍店Ztore請緊人
econ應該用好多statistic models，有好多由econ相關domain出黎 好有用既stat工具 係DS中都有用，好似t-test個啲
其實每個academic domain都會有機會用到DS/AI/ML, 因為ML只係一套方法同工具
其實每個academic domain都會有機會用到DS/AI/ML, 因為ML只係一套方法同工具
traditionally, econometrics focuses on estimation and hypothesis test, evaluate the estimator by its small sample and large sample properties.
but the situation is changing...
Machine Learning Methods That Economists Should Know About by Athey and Imbens
https://www.annualreviews.org/doi/full/10.1146/annurev-economics-080217-053433
Data science and machine learning focuses on prediction and focuses less on the asymptotic distribution of the estimator. (However, there are exceptions, the large sample properties of LASSO regression has been derived).
but the situation is changing...
Machine Learning Methods That Economists Should Know About by Athey and Imbens
https://www.annualreviews.org/doi/full/10.1146/annurev-economics-080217-053433
Data science and machine learning focuses on prediction and focuses less on the asymptotic distribution of the estimator. (However, there are exceptions, the large sample properties of LASSO regression has been derived).
係咪會開個谷？
有個谷會容易凝聚有心人，學習路途上有個伴，會好得多。
有個谷會容易凝聚有心人，學習路途上有個伴，會好得多。
一路有睇開reference book 
咁睇黎操下leetcode 好似都有用如果for揾工?
其實鐘意讀嘅野好多不過始終要面對現實
黎緊就grad 真係要揾返啲practical skill/knowledge 裝備下自己
咁睇黎操下leetcode 好似都有用如果for揾工?其實鐘意讀嘅野好多
不過始終要面對現實 黎緊就grad 真係要揾返啲practical skill/knowledge 裝備下自己
Real thank you ching我原本對行情/將來做野係點真係冇咩idea
話說俾你講中咗近排真係開始睇緊PRML一開始嘅bayesian approach intro 真係幾正
係與此同時唔知點砌portfolio... 我都有係咁八下啲ds channel 真係對fg 入行好悲觀
咁而家睇黎re pick up 返cs 個邊好似expected value 最好
btw measure theory嘅話youtube 有個巴西佬教得幾好
不過就唔會講到去probability
我原本對行情/將來做野係點真係冇咩idea話說俾你講中咗近排真係開始睇緊PRML
一開始嘅bayesian approach intro 真係幾正係與此同時唔知點砌portfolio... 我都有係咁八下啲ds channel 真係對fg 入行好悲觀
咁而家睇黎re pick up 返cs 個邊好似expected value 最好
btw measure theory嘅話youtube 有個巴西佬教得幾好
不過就唔會講到去probability幾錢人工到?
同意啊, 但係我覺得as學習谷唔好一次過太多人
4-6人一個小組一齊學一齊做最好
幾十人既話基本上你無咩動力去做/學呢件事
4-6人一個小組一齊學一齊做最好
幾十人既話基本上你無咩動力去做/學呢件事