Watson Studio ๋‚ด Community์—์„œ ์ œ๊ณตํ•˜๋Š” Single Convolution layer on MNIST ์˜ˆ์ œ ๋ชจ๋ธ์„ Modeler Flows ๋ฅผ ์‚ฌ์šฉํ•˜์—ฌ ์ƒ์„ฑํ•ด ๋ณด๊ฒ ์Šต๋‹ˆ๋‹ค.

1.IBM Cloud์—์„œ Watson Studio Service๋ฅผ ์ƒ์„ฑํ•ฉ๋‹ˆ๋‹ค.

2.IBM Cloud์—์„œ Object Storage Service๋ฅผ ์ƒ์„ฑํ•ฉ๋‹ˆ๋‹ค.
Watson Studio๋Š” Object Storage์— ๋ชจ๋ธ ๋“ฑ์˜ ์ •๋ณด๋ฅผ ์ €์žฅํ•˜๊ธฐ ๋•Œ๋ฌธ์— ๋ฐ˜๋“œ์‹œ ํ•„์š”ํ•œ ์„œ๋น„์Šค ์ž…๋‹ˆ๋‹ค.

3.New Project ๋ฅผ ์ƒ์„ฑํ•ฉ๋‹ˆ๋‹ค.

1)New Project ๋ฅผ ์„ ํƒํ•ฉ๋‹ˆ๋‹ค.

2)Project tile : Complete (Modeler, Watson Tools ๋“ฑ ๋ชจ๋“  Tools์„ ํฌํ•จํ•˜๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค.)

3)Project Name : Neural Network Project

4)Object Storage : ์œ„์—์„œ ์ƒ์„ฑํ•œ object storage ์„ ํƒ

5)์ƒ์„ฑ๋œ Project๋ฅผ ํ™•์ธํ•ฉ๋‹ˆ๋‹ค.

4.Neural Network Model์—์„œ ์‚ฌ์šฉํ•  Sample Data๋ฅผ IBM Cloud Object Storage์— uploadํ•ฉ๋‹ˆ๋‹ค.

1)MNIST.ZIP ์„ ๋‹ค์šด๋กœ๋“œ ํ•˜์—ฌ ์••์ถ•์„ ํ’‰๋‹ˆ๋‹ค.

2)Neural Network Project ์—์„œ ์‚ฌ์šฉํ•˜๊ณ  ์žˆ๋Š” Cloud Object Storage ๋กœ ์ด๋™ํ•ฉ๋‹ˆ๋‹ค.

3)๋ฒ„ํ‚ท์„ ์ƒ์„ฑํ•ฉ๋‹ˆ๋‹ค.

4)์˜ค๋ธŒ์ ํŠธ ์ถ”๊ฐ€๋ฅผ ํด๋ฆญํ•˜์—ฌ ํŒŒ์ผ์„ ์—…๋กœ๋“œ ํ•ฉ๋‹ˆ๋‹ค.

5)์—…๋กœ๋“œ๋œ ํŒŒ์ผ์„ ํ™•์ธํ•ฉ๋‹ˆ๋‹ค.

5.Modeler Flows Tool์„ ์‚ฌ์šฉํ•˜์—ฌ Model์„ ์ƒ์„ฑํ•ฉ๋‹ˆ๋‹ค.

1)์ƒ์„ฑํ•œ Neural Network Project์—์„œ Modeler Flows ์šฐ์ธก์˜ +New flow๋ฅผ ์„ ํƒํ•ฉ๋‹ˆ๋‹ค.

2)Modeler์—์„œ From example ํƒญ์„ ํด๋ฆญํ•ฉ๋‹ˆ๋‹ค.
Single Convolution layer on MNIST ๋ฅผ ์„ ํƒํ•ฉ๋‹ˆ๋‹ค.

3)์ƒ์„ฑ๋œ Model Flow์—์„œ Input ๋ฐ์ดํ„ฐ๋ฅผ ์„ ํƒํ•ฉ๋‹ˆ๋‹ค.

4)Cloud Object Storage ์— ์—…๋กœ๋“œํ•œ ๋ฐ์ดํ„ฐ์…‹๊ณผ ์—ฐ๊ฒฐํ•˜๊ธฐ ์œ„ํ•œ Connection์„ ์ƒ์„ฑํ•˜๊ณ  bucket๋ฅผ ์„ ํƒํ•ฉ๋‹ˆ๋‹ค.

5)bucket๋‚ด์—์„œ trian, test, validate ์šฉ ํŒŒ์ผ์„ ๊ฐ๊ฐ ์„ ํƒํ•ฉ๋‹ˆ๋‹ค.

6)์ƒ์„ฑํ•œ ๋ชจ๋ธ Flow๋ฅผ Watson Machine Learning ์„œ๋น„์Šค์— Deploy ํ•˜์—ฌ ํŠธ๋ ˆ์ด๋‹ ์‹œํ‚ค๊ธฐ ์œ„ํ•ด Publish ๋ฅผ ํด๋ฆญํ•ฉ๋‹ˆ๋‹ค.

7)Watson Machine Learning์— Deployํ•  ์ด๋ฆ„ ์ฆ‰, Publish Training Definition ์ด๋ฆ„์„ ์ •์˜ํ•ฉ๋‹ˆ๋‹ค.

9)Deploy ๋˜๋Š” WML Instance๋ฅผ ์„ ํƒํ•ฉ๋‹ˆ๋‹ค. WML Instance ๋ชฉ๋ก์ด ๋ณด์ด์ง€ ์•Š์„ ๊ฒฝ์šฐ ํ•˜๋‹จ ์šฐ์ธก์˜ Setting ์„ ํƒํ•ฉ๋‹ˆ๋‹ค.
Neural Network Project์˜ Setting ๋กœ ํ™”๋ฉด์ด ์ „ํ™˜ ๋ฉ๋‹ˆ๋‹ค. ์Šคํฌ๋กค ๋‹ค์šดํ•˜์—ฌ Associated Service๋ฅผ ํด๋ฆญํ•ฉ๋‹ˆ๋‹ค.

10) Watson Service๋ฅผ ์„ ํƒํ›„ Watson Machine Learning์„ ์„ ํƒํ•ฉ๋‹ˆ๋‹ค.

11)Neural Network Project ์—์„œ ์‚ฌ์šฉ๋˜๋Š” WML ์„œ๋น„์Šค๋ฅผ ํ™•์ธํ•ฉ๋‹ˆ๋‹ค.

12)๋‹ค์‹œ Modeler Flows ์˜ Single Convolution layer on MNIST ๋ฅผ ์„ ํƒํ•˜์—ฌ Publish๋ฅผ ํด๋ฆญํ•ฉ๋‹ˆ๋‹ค. Publishํ•  WML์„ ์„ ํƒํ•ฉ๋‹ˆ๋‹ค.

13)Training Definition์ด ์„ฑ๊ณต์ ์œผ๋กœ publish๋˜์—ˆ๋Š”์ง€ ์™ผ์ชฝ ์ƒ๋‹จ ๋ฉ”์‹œ์ง€๋ฅผ ํ†ตํ•ด ํ™•์ธํ•ฉ๋‹ˆ๋‹ค.

 

๋‹ค์Œ ํšŒ์—์„œ๋Š” Neural Network Experiments๋ฅผ ์‚ฌ์šฉํ•˜์—ฌ Training Definition์„ ํ•™์Šต, ํ…Œ์ŠคํŠธ, ๊ฒ€์ฆํ•ด ๋ณด๋„๋ก ํ•˜๊ฒ ์Šต๋‹ˆ๋‹ค.

์›๋ณธ: IBMยฎ Watson Studio – ML/DL made easy

ํ† ๋ก  ์ฐธ๊ฐ€

์ด๋ฉ”์ผ์€ ๊ณต๊ฐœ๋˜์ง€ ์•Š์Šต๋‹ˆ๋‹ค. ํ•„์ˆ˜ ์ž…๋ ฅ์ฐฝ์€ * ๋กœ ํ‘œ์‹œ๋˜์–ด ์žˆ์Šต๋‹ˆ๋‹ค.