[MiniQ] Machine Learning with Amazon SageMaker
- 1. JULY 19, 2018 ALIAKSEI STATUT Machine Learning with Amazon SageMaker
- 2. 2 仍亠从亠亶 弌舒 Software Engineer 仗 舒弍仂 弍仂仍亠亠 3- 仍亠 弌舒舒ﾑ 仆亠 亰舒亳从仍亳于舒 仆舒 从仂仆从亠仆 亠仆仂仍仂亞亳 丐 亊?
- 3. 3 丼 ? 舒从仂 仂 Machine Learning1 AWS SageMaker2 亳仆亶 仂仗3
- 4. 4 舒亳仆仆仂亠 仂弍亠仆亳亠 (Machine Learning) 仂弍亳仆亶 仗仂亟舒亰亟亠仍 亳从于亠仆仆仂亞仂 亳仆亠仍仍亠从舒, 亳亰舒ﾑ亳亶 仄亠仂亟 仗仂仂亠仆亳 舒仍亞仂亳仄仂于, 仗仂仂弍仆 仂弍舒, 亳 亳 仗亳仄亠仆亠仆亳亠.
- 5. 5 丐乘 丼
- 6. 6 弌弌个丶亊
- 7. 7 丼 弌弌丐亊 弌弌
- 8. 8 丼 弌丐丶 (丕丼亊 丕丼丐亊)
- 9. 9 丼 亊
- 10. 10 亅丐乘 丐 弌弍仂 亟舒仆仆 仂亟亞仂仂于从舒 亟舒仆仆 弍仂 仄仂亟亠仍亳 亳 亠仆亳仂于从舒 亠仆从舒 亳 仂仆从舒 仆舒仂亶从舒 仗仂仍亰仂于舒仆亳亠 亳 仍亠仆亳亠
- 11. 11 1. 舒仍亶 仂弍亠仄 仂弍舒ﾑ亠亶 于弍仂从亳 2. 亠从仂亠从仆仂 于仂亟仆 亟舒仆仆 3. 亠亠仂弍亠仆亳亠 弌乘 乘 ML
- 12. 12 AWS SageMaker - 仂 仗舒于仍磳仄舒 仗仍舒仂仄舒 亟仍 弍仂亞仂 亳 仗仂仂亞仂 仂亰亟舒仆亳, 仂弍亠仆亳 亳 舒亰于亠于舒仆亳 仄仂亟亠仍亳 仄舒亳仆仆仂亞仂 仂弍亠仆亳 于 仍ﾐ頴笑 仄舒舒弍亠.
- 13. 13 1. 仆仄亠仆 亟仍 弍仂舒 亟舒仆仆 2. 亳亰舒仍亳亰舒亳 亟舒仆仆 Jupyter Notebook 3. 仂亠仆仆亠 ML 舒仍亞仂亳仄 4. 舒亰于亠于舒仆亳亠 亳 A/B 亠亳仂于舒仆亳亠 5. 于仂仄舒舒弍亳仂于舒仆亳亠 AWS SAGEMAKER 丕乂丐 亂?
- 15. 15 弌弍仂 亟舒仆仆 仂亟亞仂仂于从舒 亟舒仆仆 弍仂 仄仂亟亠仍亳 亳 亠仆亳仂于从舒 亠仆从舒 亳 仂仆从舒 仆舒仂亶从舒 仗仂仍亰仂于舒仆亳亠 亳 仍亠仆亳亠 - S3 - Redshift - DynamoDB - RDS Jupyter Notebook SageMaker Training jobs - Jupyter Notebook - Hyperparameter tuning jobs - Endpoints - Cloudwatch
- 16. 16 1. Notebook instances 2. Training jobs 3. Hyperparameter tuning jobs 4. Models 5. Endpoint configurations 6. Endpoints 丐乘
- 17. 17 丐 乘丱
- 18. 18 丕丶亊 乘丱
- 19. 19 1. 弌 仗仂仄仂 SageMaker API 2. 弌 仗仂仄仂 UI 弍舒 舒仍亞仂亳仄 仗亠亟亠仍亳 舒亰仄亠舒仄亳 于亳仍亳亠仍仆 仄仂仆仂亠亶 丕从舒亰舒 仗舒舒仄亠 舒仍亞仂亳仄舒 舒亟舒 亳仂仆亳从 亟舒仆仆 ( Amazon S3) 舒仗亳 亠仆亳仂于从 弌 丐丼 亊
- 20. 20 1. 弍舒 舒仍亞仂亳仄 2. 弍舒 亟亳舒仗舒亰仂仆 仗舒舒仄亠仂于 3. 弍舒 仄亠亳从 4. 丕从舒亰舒 亳仂仆亳从 亟舒仆仆 (Amazon S3) 5. 舒亟舒 仂亞舒仆亳亠仆亳亠 亟仍 于亳仍亳亠仍仆 亠仂于 丐丐丼弌亊 弌丐 (HYPERPARAMETER TUNING)
- 21. 21 1. 仗仂仍亰仂于舒仆亳亠 仄仂亟亠仍亳 亠亠亰 HTTPS 亠 2. 仂亰仄仂亢仆仂 仂舒仆亠仆亳亠 仗仂仍亰仂于舒亠仍从亳 亰舒仗仂仂于 亟仍 仗仂仍亠亟ﾑ亠亞仂 仍亠仆亳 仄仂亟亠仍亳 3. 舒仂亶从舒 A/B 亠亳仂于舒仆亳 亟仍 亠舒 4. 于仂仄舒亳亠从仂亠 仄舒舒弍亳仂于舒仆亳亠 弌亂
- 22. 22 1. Linear Learner 2. XGBoost Algorithm 3. Image Classification Algorithm 4. K-Means Algorithm 5. ... 6. + 仂亰仄仂亢仆仂 亳仗仂仍亰仂于舒 于仂亳 弌丐乘 丐乘
- 23. 23 1. AWS SageMaker SDK 2. 仆亠亞舒亳 Apache Spark 3. 亳弍从舒 仆舒仂亶从舒 弍亠亰仂仗舒仆仂亳 4. 仂仆亳仂亳仆亞 丕 乘 弌丐
- 24. 24 丶亠仆舒 (https://aws.amazon.com/sagemaker/pricing) 弌丐丐 仂仍-于仂 舒仂于 于 仄亠 丐亳仗 亠于亠舒 亟仍 亠仆亳仂于从亳 丐亳仗 亠于亠舒 亟仍 仂亳仆亞舒 弌仂亳仄仂 舒舒 亠亞仂 4 ml.m4.xlarge $1.12 $4.48 24*31*3 = 2232 ml.t2.medium $0.065 $145.08
- 25. 25 1. 舒亟舒舒: 亠亟从舒亰舒 CTR (Click through rate) 亟仍 亠从仍舒仄仆 弍仍仂从舒 仆舒 舒仆亳亠 2. 舒舒仄亠: 舒仗仂仍仂亢亠仆亳亠 弍仍仂从舒 丕仂亶于仂 仗仂仍亰仂于舒亠仍 丱舒舒从亠亳亳从亳 仗仂仍亰仂于舒亠仍 丱舒舒从亠亳亳从亳 亠从仍舒仄亳亠仄仂亞仂 仂于舒舒 3. 亠亰仍舒: 仄仂亟亠仍 仆亠亟舒仆舒, ctr 亳 仗舒舒仄亠 仆亠 亳仄亠ﾑ 亳仍仆仂亶 亰舒于亳亳仄仂亳 仄亠亢亟 仂弍仂亶 丼乘 乘丐
- 26. 26 仂仗仂?
Editor's Notes
- #3: P37 divide into 2 slides