[MLS-C01] [Modeling] Train machine learning models

Training the model is part of the iterative model improvement cycle

Training & Inference Instances

Training and inference instances for SageMaker built-in algorithms

The Steps of Training ML Model

1. Gather/Engineer

• Gather/engineer data into your dataset

2. Randomize & Split

• Randomize the dataset
• Split the dataset into train and test datasets

3. Choose Algorithm

• Choose best algorithm

4. Load Container & Define Hyperparameters

• Load container for chosen model
• Manage compute capacity
• Create an instance of chosen model
• Define the model’s hyperparameter values

5. Train the Model

• Train the model

6. Deploy the Model

Deploy in one of two ways

• Persistent endpoint to get one prediction at a time: SageMaker Hosting Services
• Get predictions for an entire dataset: SageMaker Batch Transform

SageMaker Hosting Services

SageMaker provides an HTTPS endpoint, making your model available for inference requests

Three steps:

1. Create model in SageMaker
2. Create an endpoint configuration
3. Create HTTPS endpoint

SageMaker Batch Transform

SageMaker batch transform provides inferences for an entire dataset

Three steps:

1. Create batch transform job using trained model and dataset
2. Run the batch transform job
3. SageMaker saves in a results S3 bucket

Labs

• train-deploy-evaluate-model.ipynb
• adult_census_clean.csv
• SageMaker Hosting Services
• XGBoost Algorithm - Amazon SageMaker

本文由 Oscaner 创作, 采用 知识共享署名4.0 国际许可协议进行许可
本站文章除注明转载/出处外, 均为本站原创或翻译, 转载前请务必署名

• Previous
  [MLS-C01] [Modeling] Modeling Concepts
• Next
  [MLS-C01] [Modeling] Perform automatic model tuning