Openscale์€ ๋จธ์‹ ๋Ÿฌ๋‹ ๋ชจ๋ธ์„ ๋น„์ฆˆ๋‹ˆ์Šค ์–ดํ”Œ๋ฆฌ์ผ€์ด์…˜์—์„œ ์‚ฌ์šฉํ•˜๋Š” ํ™˜๊ฒฝ์—์„œ ๋จธ์‹ ๋Ÿฌ๋‹ ๋ชจ๋ธ์˜ ์˜ˆ์ธก ๊ฒฐ๊ณผ๊ฐ’์ด ์ •ํ™•ํ•˜๊ฒŒ ์˜ˆ์ธกํ•˜๊ณ  ์žˆ๋Š”์ง€, ๋น„์ฆˆ๋‹ˆ์Šค์ ์œผ๋กœ ๊ณตํ‰ํ•œ ์˜ˆ์ธก์ด ํ•„์š”ํ•œ ์†์„ฑ(์˜ˆ:์„ฑ๋ณ„, ๋‚˜์ด, ์ธ์ข…)์— ๋Œ€ํ•ด ํŽธํ–ฅ(Bias)๊ฐ€ ์—†๋Š”์ง€, ๋‚˜์•„๊ฐ€ ์™œ ์ด๋Ÿฌํ•œ ๊ฒฐ๊ณผ๊ฐ€ ๋‚˜์™”๋Š”์ง€ ํŠธ๋žœ์žญ์…˜ ๋‹จ์œ„๋กœ ์„ค๋ช…ํ•  ์ˆ˜ ์žˆ๋Š” ๊ธฐ๋Šฅ์„ ์ œ๊ณตํ•˜๋Š” ๋จธ์‹ ๋Ÿฌ๋‹ ๋ชจ๋ธ ๊ด€๋ฆฌ ๋ฐ ์šด์˜ ํ”Œ๋žซํผ์ž…๋‹ˆ๋‹ค.

๋ณด๋‹ค ์‹ค์ œ์ ์ธ ์ดํ•ด๋ฅผ ์œ„ํ•ด ์ž๋™์ฐจ ๋ Œํƒˆ ๋ฐ์ดํ„ฐ๋ฅผ ์‚ฌ์šฉํ•˜์—ฌ ๋ Œํƒˆ ์„œ๋น„์Šค์— ๋งŒ์กฑํ•˜์ง€ ๋ชปํ•œ ๊ณ ๊ฐ๋“ค์„ ๋Œ€์ƒ์œผ๋กœ ํ”ฝ์—…์„œ๋น„์Šค ํ˜น์€ ๋ฐ”์šฐ์ฒ˜ ๋“ฑ์˜ ์ถ”์ฒœ ์•ก์…˜(Action)์„ ์˜ˆ์ธกํ•˜๋Š” ๋ชจ๋ธ์„ Openscale์„ ์ ์šฉํ•˜๋Š” ํŠœํ† ๋ฆฌ์–ผ์„ ์ง„ํ–‰ํ•ด ๋ณด๋„๋ก ํ•˜๊ฒ ์Šต๋‹ˆ๋‹ค.

Openscale์€ ๋จธ์‹ ๋Ÿฌ๋‹ ๋ชจ๋ธ์„ ๋ชจ๋‹ˆํ„ฐ๋งํ•˜๋Š” ์„œ๋น„์Šค๊ธฐ ๋•Œ๋ฌธ์— Build ๋ฐ Deployed ๋จธ์‹ ๋Ÿฌ๋‹ ๋ชจ๋ธ์ด ๋ฐ˜๋“œ์‹œ ํ•„์š”ํ•ฉ๋‹ˆ๋‹ค. ํŠœํ† ๋ฆฌ์–ผ์—์„œ๋Š” ๊ณต๊ฐœ Spark ๊ธฐ๋ฐ˜์˜ CARS4U-Action-Recommendation ๋ชจ๋ธ์„ IBM Watson Studio Service ๋ฅผ ํ†ตํ•ด Buildํ•˜๊ณ  IBM Watson Machine Learning Service์— Deploy ํ•˜๊ณ  ํ•ด๋‹น CARS4U-Action-Recommendation ๋ชจ๋ธ์„ Openscale์— ์—ฐ๊ฒฐํ•˜์—ฌ ๋ชจ๋‹ˆํ„ฐ๋ง Feature(์†์„ฑ) ๋ฐ threshold ๋“ฑ์„ ์ •ํ•ด ๋ณด๋„๋ก ํ•˜๊ฒ ์Šต๋‹ˆ๋‹ค.

ํŠœํ† ๋ฆฌ์–ผ์— ์‚ฌ์šฉ๋˜๋Š” ์ฃผ์š” ์„œ๋น„์Šค

1. Watson Studio
2. Watson Machine Learning
3. Watson Openscale
4. DB2 (optional)

Openscale ํŠœํ† ๋ฆฌ์–ผ๋กœ Watson Studio ์™€ Watson Machine Learning ์„œ๋น„์Šค๋ฅผ ์‚ฌ์šฉํ•˜๋Š” ๋ถ€๋ถ„์€ ๊ฐ„๋‹จํ•˜๊ฒŒ ์ง„ํ–‰ํ•˜๊ฒ ์Šต๋‹ˆ๋‹ค.

๋‹จ๊ณ„ 1 – ๋จธ์‹ ๋Ÿฌ๋‹ ๋ชจ๋ธ Build ํ•˜๊ธฐ

1.๋จธ์‹ ๋Ÿฌ๋‹์— ์‚ฌ์šฉํ•  ํ•™์Šต๋ฐ์ดํ„ฐ๋ฅผ ํ™•์ธํ•ด ๋ณด๊ฒ ์Šต๋‹ˆ๋‹ค. https://github.com/watson-developer-cloud/doc-tutorial-downloads/blob/master/ai-openscale/car_rental_training_data.csv

ํ•™์Šต๋ฐ์ดํ„ฐ๋Š” Gender(์„ฑ๋ณ„), Children(์ž๋…€์ˆ˜),Age(์—ฐ๋ น),Car_Owner(์ฐจ๋Ÿ‰๋ณด์œ ์—ฌ๋ถ€),Satisfaction(์„œ๋น„์Šค๋งŒ์กฑ๋„),Action(์ถ”์ฒœ์„œ๋น„์Šค) ๋“ฑ์˜ ์†์„ฑ์œผ๋กœ ๊ตฌ์„ฑ๋˜์–ด ์žˆ์Šต๋‹ˆ๋‹ค.

2.ํ•™์Šต ๋ฐ์ดํ„ฐ๋ฅผ DB2์— ์˜ฌ๋ฆฌ๊ณ  ์ด๋ฅผ Watson Studio์˜ Asset์œผ๋กœ Connectํ•ฉ๋‹ˆ๋‹ค. DB2 ์„œ๋น„์Šค๋ฅผ ์‚ฌ์šฉํ•˜์ง€ ์•Š์„ ๊ฒฝ์šฐ๋Š” csv ํŒŒ์ผ์„ Watson Studio์˜ Asset์— ๋ฐ”๋กœ upload ํ•˜์…”๋„ ๋ฉ๋‹ˆ๋‹ค.

3. Watson Studio์˜ Data Profiling ๊ธฐ๋Šฅ์„ ์‚ฌ์šฉํ•˜์—ฌ ๊ฐ„๋‹จํ•œ ํ†ต๊ณ„ ์ •๋ณด ๋“ฑ์„ ํ†ตํ•ด ๋จธ์‹ ๋Ÿฌ๋‹ ๋ชจ๋ธ์˜ ์‹œ์ž‘ ๋‹จ๊ณ„์ธ ๋ฐ์ดํ„ฐ ํƒ์ƒ‰์„ ํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.

4. Watson Studio์— CARS4U-Action-Recommendation-spark.ipynb์„ Jupyter Notebook์œผ๋กœ importํ•˜์—ฌ ์‹คํ–‰ํ•ฉ๋‹ˆ๋‹ค. https://github.com/watson-developer-cloud/doc-tutorial-downloads/blob/master/ai-openscale/CARS4U-Action-Recommendation-Spark.ipynb

๋‹จ๊ณ„ 2 – ๋จธ์‹ ๋Ÿฌ๋‹ ๋ชจ๋ธ Deploy

1. CARS4U-Action-Recommendation-spark ๋ชจ๋ธ์„ Watson Studio์˜ Notebook ์ƒ์—์„œ Watson Machine Learning ์— Deploy ํ•˜๋ฉด ์•„๋ž˜์™€ ๊ฐ™์ด Model Name, Deployment ๊ด€๋ จ์ •๋ณด๋ฅผ ๋ฆฌํ„ดํ•ด ์ค๋‹ˆ๋‹ค.

์ด์ œ Openscale์— ์—ฐ๋™ํ•  ๋จธ์‹ ๋Ÿฌ๋‹ ๋ชจ๋ธ์ด ์ค€๋น„๋˜์—ˆ์Šต๋‹ˆ๋‹ค. ๋ณธ๊ฒฉ์ ์œผ๋กœ Openscale ํŠœํ† ๋ฆฌ์–ผ์„ ์‹œ์ž‘ํ•ด ๋ณด๊ฒ ์Šต๋‹ˆ๋‹ค.

๋‹จ๊ณ„ 3 – ๋จธ์‹ ๋Ÿฌ๋‹ ๋ชจ๋ธ์„ Openscale์— ์—ฐ๋™ํ•˜๊ธฐ

1.OpenScale ์„œ๋น„์Šค์˜ Tool์„ Openํ•ฉ๋‹ˆ๋‹ค.

๋””ํดํŠธ ๋ฐ์ดํ„ฐ๋ฒ ์ด์Šค๋กœ Postgre๋ฅผ ์„ ํƒํ•ฉ๋‹ˆ๋‹ค. schema๋Š” public์œผ๋กœ ์„ ํƒํ•ฉ๋‹ˆ๋‹ค. ๋ฐ์ดํ„ฐ๋ฒ ์ด์Šค๋Š” ์•ž์— ์ƒ์„ฑํ•œ ๋ชจ๋ธ Deployment ๊ฒฐ๊ณผ๋ฅผ Openscale์—๋„ ์ €์žฅํ•˜๊ณ  ์žฌํ•™์Šต์„ ์œ„ํ•œ ๋ฐ์ดํ„ฐ๋ฅผ ์ €์žฅํ•˜๋Š” ์šฉ๋„๋กœ ์‚ฌ์šฉ๋ฉ๋‹ˆ๋‹ค.

2.Openscale์—์„œ ๋ชจ๋‹ˆํ„ฐ๋งํ•œ ๋ชจ๋ธ์„ ์„ ํƒํ•˜๊ธฐ ์œ„ํ•ด ํ•ด๋‹น ๋ชจ๋ธ์ด Deploy๋œ ํ™˜๊ฒฝ์„ ์„ ํƒํ•ฉ๋‹ˆ๋‹ค.
์ €ํฌ๋Š” ์•ž๋‹จ๊ณ„์—์„œ Watson Machine Learning์— ๋ชจ๋ธ์„ Deployํ–ˆ๊ธฐ ๋•Œ๋ฌธ์— Watson Machine Learning์„ ์„ ํƒํ•ฉ๋‹ˆ๋‹ค.

3.Watson Machine Learning์— Deploy๋œ ๋ชจ๋ธ์ค‘ ๋ชจ๋‹ˆํ„ฐ๋ง ๋Œ€์ƒ ๋ชจ๋ธ์„ ์„ ํƒํ•ฉ๋‹ˆ๋‹ค.
์ €ํฌ๋Š” ์•ž๋‹จ๊ณ„์—์„œ 1๊ฐœ์˜ ๋ชจ๋ธ์„ 1๋ฒˆ Deployํ–ˆ๊ธฐ ๋•Œ๋ฌธ์— Deploy๋œ ๋ชจ๋ธ์„ 1๊ฐœ๋งŒ ์กด์žฌํ•ฉ๋‹ˆ๋‹ค.

4.์ง€๊ธˆ๊นŒ์ง€๋Š” ์šฐ๋ฆฌ๊ฐ€ ๋ชจ๋‹ˆํ„ฐ๋งํ•  ๋ชจ๋ธ์˜ Deployment๋ฅผ Openscale์— ์—ฐ๊ฒฐํ•˜๋Š” ๊ธฐ๋ณธ ์„ค์ • ์ž‘์—…์ด์—ˆ์Šต๋‹ˆ๋‹ค. ์ด๋ฅผ ํ™•์ธํ•˜๊ณ  ๋‹ค์Œ ๋‹จ๊ณ„๋กœ ๋„˜์–ด๊ฐ‘๋‹ˆ๋‹ค.

5.์œ„์—์„œ Openscalel์— ์„ค์ •ํ•œ Deployment๋ฅผ ์„ ํƒํ•ฉ๋‹ˆ๋‹ค.

6.์•„๋ž˜ ๊ทธ๋ฆผ์ฒ˜๋Ÿผ Openscale์—์„œ๋Š” Accuracy, Fairness, Explainability ์˜ 3๊ฐ€์ง€ ์˜์—ญ์— ๋Œ€ํ•ด ๋จธ์‹ ๋Ÿฌ๋‹ ๋ชจ๋ธ์„ ์šด์˜ ๋ฐ ๋ชจ๋‹ˆํ„ฐ๋ง ํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.

๋ณธ๊ฒฉ์ ์ธ Openscale์˜ 3๊ฐ€์ง€ ๋ชจ๋‹ˆํ„ฐ๋ง ๋Œ€์ƒ ์˜์—ญ์— ๋Œ€ํ•ด์„œ๋Š” ๋‘๋ฒˆ์งธ ๊ธ€์—์„œ ์ง„ํ–‰ํ•˜๋„๋ก ํ•˜๊ฒ ์Šต๋‹ˆ๋‹ค.