ABOUT ME

-

Today
-
Yesterday
-
Total
-
  • [Machine Learning] Basic Of Machine Learning
    ๐ŸณDev/Machine Learning 2022. 1. 15. 15:20

    ๋ชจ๋‘๋ฅผ ์œ„ํ•œ ๋”ฅ๋Ÿฌ๋‹ (๊น€์„ฑํ›ˆ)


    1. Machine Learning

    ๊ธฐ๊ณ„ํ•™์Šต์€ ์ˆ˜์ง‘๋œ ๋ฐ์ดํ„ฐ(๋กœ๊ทธ, ์Œ์„ฑ, ํ…์ŠคํŠธ, ์ด๋ฏธ์ง€) ์†์—์„œ ํŒจํ„ด์„ ์ฐพ์•„๋‚ด๋„๋ก ํ•™์Šต๋˜๋Š” ์•Œ๊ณ ๋ฆฌ์ฆ˜ ๋ชจ๋ธ์ด๋‹ค. ๊ธฐ๊ณ„ํ•™์Šต์€ ๊ฒฐ๊ณผ๊ฐ€ ์ฃผ์–ด์ง„ ๋ฐ์ดํ„ฐ๋กœ ํ•™์Šต์„ ํ•˜๋Š” ์ง€๋„ ํ•™์Šต๊ณผ ์ž์œจ์ ์ธ ํ•™์Šต์„ ํ•˜๋Š” ๋น„์ง€๋„ ํ•™์Šต์œผ๋กœ ๋‚˜๋‰œ๋‹ค.

    1. Supervised Learning โ˜†
       : labeled data, training set์„ ์ด์šฉํ•ด ํ•™์Šตํ•˜๋Š” ๋ฐฉ๋ฒ•
    2. UnSupervised Learning
       : unlabeled data๋ฅผ ์ด์šฉํ•ด ํ•™์Šตํ•˜๋Š” ๋ฐฉ๋ฒ•

    ์šฐ๋ฆฌ๋Š” ์ง€๋„ํ•™์Šต์„ ํ†ตํ•ด, ๊ธฐ๊ณ„ํ•™์Šต ๊ณต๋ถ€๋ฅผ ์ง„ํ–‰ํ•œ๋‹ค.

    2. Type of Supervised Learning

    ์ง€๋„ํ•™์Šต์˜ ์ข…๋ฅ˜๋Š” ํฌ๊ฒŒ ๋ถ„๋ฅ˜(Classification) ๋ฌธ์ œ์™€ ํšŒ๊ท€(Regression) ๋ฌธ์ œ๋กœ ๋‚˜๋‰œ๋‹ค. ์ฃผ์–ด์ง„ ์ž…๋ ฅ ๋ฐ์ดํ„ฐ์—์„œ ๋ถ„๋ฅ˜๋Š” ๋ฏธ๋ฆฌ ์ •์˜ ๋œ ์—ฌ๋Ÿฌ ํด๋ž˜์Šค์—์„œ ์–ด๋””์— ์†ํ•˜๋Š”์ง€ ์˜ˆ์ธกํ•˜๊ณ , ํšŒ๊ท€๋Š” ์—ฐ์†์ ์ธ ์ˆซ์ž๋ฅผ ์˜ˆ์ธกํ•œ๋‹ค. ์–ด๋–ค ๋ฌธ์ œ๋ฅผ ๋‹ค๋ฃฐ ๋•Œ, ์ ์ ˆํ•œ ํ•™์Šต์„ ์ฐพ๋Š” ๊ฒƒ์€ ๋งค์šฐ ์ค‘์š”ํ•˜๋‹ค.

    1. ์„ ํ˜• ํšŒ๊ท€ Linear Regression
       : ์—ฐ์†ํ•œ ๋ณ€์ˆ˜๋“ค์— ๋Œ€ํ•ด ๋‘ ๋ณ€์ˆ˜ ์‚ฌ์ด์˜ ๋ชจํ˜•์„ ๊ตฌํ•˜๊ณ  ์ ํ•ฉ๋„๋ฅผ ์ธก์ •
       ex) ๊ณต๋ถ€ํ•œ ์‹œ๊ฐ„์„ ๊ธฐ์ค€์œผ๋กœ 0~100์  ์‚ฌ์ด์˜ ๊ฒฐ๊ณผ๋ฅผ ์˜ˆ์ธก
    2. ์ด์ง„ ๋ถ„๋ฅ˜ Binary Classification
       : ๋‘๊ฐ€์ง€ ๋ถ„๋ฅ˜(0 ๋˜๋Š” 1)
       ex) ๊ณต๋ถ€ํ•œ ์‹œ๊ฐ„์„ ๊ธฐ์ค€์œผ๋กœ pass or nonpass๋ฅผ ์˜ˆ์ธก
    3. ๋‹ค์ค‘ ๋ถ„๋ฅ˜ Multi-Label Classification
       : ๋‘๊ฐ€์ง€ ์ด์ƒ์˜ ๋ถ„๋ฅ˜
       ex) ๊ณต๋ถ€ํ•œ ์‹œ๊ฐ„์„ ๊ธฐ์ค€์œผ๋กœ ABCDF๋ฅผ ์˜ˆ์ธก

     

    3. TensorFlow

     : Data Flow Graph(๋ฐ์ดํ„ฐ ํ๋ฆ„ ๊ทธ๋ž˜ํ”„) + Python

    ํ…์„œํ”Œ๋กœ์šฐ๋Š” Tensor(๋‹ค์ฐจ์›๋ฐฐ์—ด)๋“ค์ด ํ๋ฅธ๋‹ค๋Š” ์˜๋ฏธ๋ฅผ ๊ฐ€์ง„๋‹ค. Node๋Š” ๋‹ค์ฐจ์› ๋ฐฐ์—ด๋กœ ์ด๋ฃจ์–ด์ง„ ์ˆ˜์น˜ ์—ฐ์‚ฐ์œผ๋กœ, edge๋Š” ๋…ธ๋“œ๋“ค์˜ ์—ฐ์‚ฐ์˜ ํ๋ฆ„์„ ์˜๋ฏธํ•œ๋‹ค.

     TensorFlow Mechanics

     1. build graph(tensor)
     2. feed data and run graph
     3. update variables in the graph and return values

     Tensor Ranks, Shapes, and Types

     1. Rank : n์ฐจ์› ๋ฐฐ์—ด
     2. Shape : ๋ฐฐ์—ด์˜ ๋ชจ์–‘์„ ์˜๋ฏธ
     3. Type : DataType(์ฃผ๋กœ float32 or int32)
     ex) t = [[1,2], [3,4], [5,6]]
       rank = 2
       shape = [3, 2]
       type = int32



Designed by Tistory.