๋ณธ๋ฌธ ๋ฐ”๋กœ๊ฐ€๊ธฐ
๋Œ€์™ธํ™œ๋™/2023 LG Aimers 3๊ธฐ

Module 4. Bayesian (๊ณ ๋ ค๋Œ€ํ•™๊ต ๊น€์žฌํ™˜)

by ์ œ๋ฃฝ 2023. 7. 4.
728x90
๋ฐ˜์‘ํ˜•

๋‚ ์งœ: 2023๋…„ 7์›” 4์ผ

Part1. Principle and Structure

1. Bayesian ์›๋ฆฌ ๋ฐ ์ž‘๋™๋ฐฉ์‹

 

 

 

 

Part 2. Estimation Algorithm

1. Joint Probablity Distribution(๊ฒฐํ•ฉ ํ™•๋ฅ  ๋ถ„ํฌ)

  • ๊ฒฐํ•ฉ ๋ถ„ํฌ๋ž€ ํ™•๋ฅ  ๋ณ€์ˆ˜๊ฐ€ ๋‘ ๊ฐœ ์ด์ƒ์ผ ๋•Œ ์—ฌ๋Ÿฌ ์‚ฌ๊ฑด์ด ๋™์‹œ์— ์ผ์–ด๋‚  ํ™•๋ฅ ์„ ๋งํ•จ.

2. Bayesian ์ถ”์ • ์•Œ๊ณ ๋ฆฌ์ฆ˜

3. Random Walk Metropolis-Hastings Algorithm

1. Posterior Distribution

  • ์ ๋ถ„ํ•˜์ง€ ์•Š๊ณ  ํ•  ์ˆ˜ ์žˆ๋Š” ๋ฐฉ๋ฒ•: Prior * Likelihood

2. Gibbs Sampling

  • ์œ„์˜ ๋ฐฉ์‹์œผ๋กœ ์•ˆ๋  ๊ฒฝ์šฐ ์‚ฌ์šฉ
  • parameter์˜ conditional ๋ถ„ํฌ๋ฅผ ์•Œ ๊ฒฝ์šฐ, Gibbs sampling ์‚ฌ์šฉ

3. Metropolis Hastings

  • conditional distribution์„ ์•Œ ์ˆ˜ ์—†์„ ๋•Œ ์‚ฌ์šฉํ•จ

1. Convergence ํ™•์ธ

2. Multiple starting points

3. Burn-in

  • ์ˆ˜๋ ดํ•œ Draw๋งŒ ์‚ฌ์šฉ

4. Autocorrelation-Acceptance rate, ACF

  • Autocorrelation Time Series ๊ฐœ๋…๊ณผ ๋น„์Šทํ•จ
  • ๋งค๋ฒˆ iteration ๋•Œ๋งˆ๋‹ค ์–ป๋Š” draw๋Š” ๋ฐ”๋กœ ์ „ ๋‹จ๊ณ„์˜ Draw๊ฐ€ ๋ฌด์Šจ ๊ฐ’์ด์—ˆ๋Š๋ƒ์— ๋”ฐ๋ผ์„œ ์ƒ๋Œ€์ ์œผ๋กœ ์˜ํ–ฅ์„ ๋ฐ›๊ฒŒ ๋˜์–ด์žˆ์Œ ⇒ Autocorrelation์ด๋ผ๊ณ  ํ•จ
  • Autocorrelation์ด ๋†’๋‹ค๋Š” ๊ฒƒ์€ rate๊ฐ€ ๋†’์€ ๊ฒƒ์„ ์˜๋ฏธํ•จ. ACF๋„ ๋†’๋‹ค๋Š” ๊ฒƒ์„ ์˜๋ฏธ

Part 3. Solving Real Problem

- ์ƒ๋žต  

728x90
๋ฐ˜์‘ํ˜•