AIF-C01 ยท ML lifecycle, MLOps and metrics
Stages of an ML pipeline
Data collection, preparation, training, evaluation, deployment and monitoring, and which AWS services support each stage.
Key points
- 1) Collect data (for example into Amazon S3).
- 2) Prepare it: clean, transform, engineer features (SageMaker Data Wrangler in Canvas).
- 3) Train a model (SageMaker training jobs, or fine-tune a foundation model).
- 4) Evaluate it on data it has not seen.
- 5) Deploy it to an endpoint or batch job.
- 6) Monitor it in production and loop back when quality drops.
Source: https://docs.aws.amazon.com/sagemaker/latest/dg/pipelines.html
Practice questions
In which pipeline stage are missing values filled in and categorical columns encoded?
- Data preparation
- Model evaluation
- Deployment
- Monitoring
Show answer
Answer: A. Data preparation
Cleaning and feature engineering belong to data preparation. Training consumes already-prepared data, evaluation measures results, and deployment serves the model.
Which activities belong to the data preparation stage?
- Imputing missing values in a dataset
- Creating new features from raw columns
- Hosting the model behind an endpoint
- Alerting when prediction quality drifts
- Running a training job on GPU instances
Show answer
Answer: A. Imputing missing values in a dataset, B. Creating new features from raw columns
Handling missing values and creating features are preparation work. Hosting an endpoint is deployment, and tracking drift is monitoring.
Once a model is deployed, the ML pipeline is finished and no further stages are needed.
Show answer
Answer: False
False. Deployed models are monitored, and drift or new data sends the team back to preparation and re-training.
Practice 8 questions on this topic in Bitzu
Short lessons, instant explanations and spaced review of what you get wrong. Free on iOS and Android.
