MLCOST-03: Identify if machine learning is the right solution
Evaluate if there are alternatives, such as a simple rule-based approach, that could do a better job than ML. Weigh the cost of adopting ML against the opportunity cost of not leaning on ML transformation. Specialized resources, such as data scientist time or model time-to-market, might be the most expensive and constrained resources. The most cost-effective hardware choice might not be cost optimized if it constrains experimentation and development speed.
Implementation plan
-
Start simple:
-
Articulate your problem.
-
Identify your data sources.
-
Think about cost involved in:
-
Designing or preparing your data for the model.
-
Data storage cost for ML.
-
Model training cost depending on the hardware. choice
-
Data labeling cost, if required.
-
Potential bias resulting in iterative model re-training leading to higher cost.
-
Potential cost of hosting the ML model.
-
Model maintenance costs.
-
-
Consider these data points to weigh the cost of adopting ML against the opportunity cost of not leaning on ML transformation.
-
-
Use Amazon SageMaker AI Autopilot and SageMaker AI Clarify to validate that ML is the right solution.
-
Baseline the solution by reviewing how the problem is solved today. If a rules-based solution is available, then use it as a baseline. Selecting a simple ML model for baselining can also be done using JumpStart or AWS Marketplace. AWS also provides many pre-built solutions with one-click deploy for most common business use cases.
-
Build a machine learning model using SageMaker AI or SageMaker AI Autopilot and compare the metrics of this solution against the baseline.
-
Use SageMaker AI Clarify to explain the model that you have built using SageMaker AI or Autopilot.
-
Identify if the ML model is performing better than your existing solution or a rules-based approach before investing on an ML-based solution.
-
Documents
