Developing ML solutions
Conclusion
The whole bakery, from plan to daily running
The story: Looking back at the bakery: you planned it around a real business need, set up one workshop for all the work, chose between buying and inventing recipes, taste-tested properly, opened in the way that fits your customers, and set up a routine that keeps quality steady over time.
In AI/AWS terms:
- ML lifecycle: from identifying the business problem to deploying and monitoring the model.
- SageMaker AI covers every step:
- Prepare data with Data Wrangler or your own scripts and notebooks.
- Build with Canvas (no code), pre-trained or built-in algorithms, supported frameworks, or custom Docker images.
- Deploy and monitor.
- Built-in algorithms cover supervised learning (classification, regression), unsupervised learning (clustering, dimension reduction, anomaly detection), image processing, time series, and NLP.
- Evaluation uses training, validation, and test sets. Underfitting means high bias, overfitting means high variance, and the goal is low bias with low variance.
- Deployment is self-hosted or managed. SageMaker offers real-time, batch, serverless, and asynchronous inference.
- MLOps extends DevOps to ML so models are deployed, monitored, and retrained repeatably.
For the exam: Be able to name the SageMaker feature for each lifecycle step and the inference option for each traffic pattern.