ch1. What is Data Science?

What is Data Science?

Data Science's concept has not been established however there are elements of Data Science and Statistics.

• Exploratory Data analysis and visualization
• Machine learning and statistics
• High - performance computing technology

Computer Science + Statistics -> Machine Learning + Data analysis + Domain Science(business) => Data Science

=> 데이터 과학자는 domain 전문가와 소통할 수 있는 능력을 학습해야한다. 이때 domain 전문가와의 소통이란 Needs를 정확하게 이해하고 결과의 insight를 전달하는 storytelling 능력을 의미한다.

 

Data Science pipline

1. Ask Interesting Questions :
   • Domain 전문가의 요구 이해하기
   • Data에서 얻을 수 있는 Insight 발굴
   • Context에서 문제 발굴
2. Get Data
3. Eplore Data
   • Plot => Find Anomalies & patterns
4. Model Data
5. Communicate and visualize results

Scientist vs Computer Scientist

Data driven => Focus on results(findings) => discover things!	Algorithm driven => focus on methods => random data to prove correctness
Understand messy natural World	Build clean virtual world
8/13 = 0.62 => Focus on Meaning	8/13 = 0.61538461538 => Focus on Accuracy
Not all True / False	All True / False

데이터 과학자는 과학자에 가깝다.

Software Developer : produce code

Data Scientist : produce insights => 가설 도출, business direction 을 높은 의사결정권자와 소통한다.

 

Data Scientist의 소양

1. Developing Curiosity
   • In domain/application 
   • Communicate with domain experts
   • develop broader perspective on the world
2. Ast good questions
   • From a given dataset what exiting things can be learned?
   • Needs의 정확한 파악
   • Dataset 활용 (어떤 데이터 사용, 접근 방식 고안)

Properties of Data

Structured vs Unstructured

Spread Sheet	tweets
EMR(Electronic Medical Record)	Movie Review
Financial Time Series	X-ray

Quantitative vs Categorical Data

• Nominal (Categorical) : 비교만 가능 =, ≠
• Ordinal : Nominal + 크기 비교 가능 =. ≠, >, <
• Quantitiative : 계산가능
  • Internal : location of zero arbitrary (+- 유의미한 정보) 날짜, 위치... =, ≠, >, <, +, -
  • Ratio : zero fixed measurement =, ≠, >, <, +, -, ×, ÷

Classification vs Regression

classification : discrete set & small label -> predict

regression : numeric quantity & continuos -> predict