[GoogleML] Hyperparameter Tuning

2023. 9. 20. 22:55ใ†ArtificialIntelligence/2023GoogleMLBootcamp

 

 

 

Tuning Process

hyperparams ๋ณ„ ์ค‘์š”๋„๊ฐ€ ๋‹ค๋ฅด๋‹ค

 

 

 

๋žœ๋คํ•˜๊ฒŒ pick!

์™œ๋ƒํ•˜๋ฉด params ๋ณ„ (์ถ• ๋ณ„) ์ค‘์š”๋„๊ฐ€ ๋‹ค๋ฅด๊ธฐ ๋•Œ๋ฌธ

์„ฌ์„ธํ•œ ์ •๋„๊ฐ€ ๋‹ฌ๋ผ์•ผ ํ•˜๋Š”๋ฐ, grid๋Š” ๋ชจ๋‘ ๋™์ผํ•˜๊ฒŒ ๋‹ค๋ฃจ๊ธฐ ๋•Œ๋ฌธ

randomํ•˜๊ฒŒ ๋ณด๋Š” ๊ฒƒ์ด ๋” ์ข‹๋‹ค 

 

 

 

3๊ฐœ ์ด์ƒ -> ๋” ํฐ ์ž…๋ฐฉ์ฒด

 

 

 

๋” ์ž‘์€ Part๋กœ ์คŒ์ธ & ๋” ๋งŽ์€ point (dense)

 

 

 

Using an Appropriate Scale to pick Hyperparameters

๋ฒ”์œ„๋ฅผ ์ฃผ๊ณ , randomํ•˜๊ฒŒ ๋ฝ‘๋Š”๋‹ค / uniformly random

 

 

 

linear๋ณด๋‹ค log scale์„ ์“ฐ๋Š” ๊ฒƒ์ด ๋” ์ ํ•ฉํ•  ์ˆ˜ ์žˆ๋‹ค

 

 

 

๋‹ค์Œ๊ณผ ๊ฐ™์ด ๋ฒ ํƒ€๊ฐ€ ๋ถ„๋ชจ์— ๋“ค์–ด๊ฐˆ ๊ฒฝ์šฐ, 

๋‹จ์ˆœํ•œ ๋ธํƒ€๊ฐ’ ์ด์ƒ์˜ ์ค‘์š”๋„๊ฐ€ ์žˆ๋‹ค (sensitivity)

 

 

 

Hyperparameters Tuning in Practice: Pandas vs. Caviar

 

 

 

์ž‘์€ setting / computational ์œผ๋กœ ํ•˜๋‚˜์˜ model์„ ํ‰๊ฐ€ vs

๋‹ค์–‘ํ•œ ๋ชจ๋ธ, ๋‹ค์–‘ํ•œ setting์„ ๋ณ‘๋ ฌ์ ์œผ๋กœ ์ฒ˜๋ฆฌ

 

 

 

ํŒ๋‹ค์‹ vs ์บ๋น„์–ด

 

 

 

 

 

'ArtificialIntelligence > 2023GoogleMLBootcamp' ์นดํ…Œ๊ณ ๋ฆฌ์˜ ๋‹ค๋ฅธ ๊ธ€

[GoogleML] Hyperparameter Tuning, Regularization and Optimization ์ˆ˜๋ฃŒ  (0) 2023.09.22
[GoogleML] Batch Normalization  (0) 2023.09.21
[GoogleML] Adam Optimizer  (0) 2023.09.20
[GoogleML] Optimization Algorithms  (0) 2023.09.20
[GoogleML] Optimization Problem  (0) 2023.09.13