MLCOST-22: Select optimal algorithms

Identify the basic machine learning paradigm that addresses your ML problem type. Basic machine learning paradigms include: supervised learning, unsupervised learning and reinforcement learning. Identify the acceptable level of tradeoff between explainability and success metrics per business requirements. Run prototypes and experiments to explore high performing algorithms. Select the optimal cost-efficient algorithms that meet all the business requirements. Improved runtime performance of a tuned algorithm within business requirements, is one step towards optimizing the cost of ML.

Implementation plan

• Adopt optimal practices

  • Start with simple algorithms, such as regression, and work towards more complex algorithms, such as deep learning, to compare the accuracy of the models. Optimize hyperparameters to determine which algorithm yields the best metrics for the business use case.

  • When selecting the optimal algorithm, run trade-off analysis between data constraints, computational performance, and maintenance efforts. For example, deep learning networks might produce more accurate results, but require more data than tree-based methods. Deep learning methods are also more difficult to maintain.

• Use AWS services

  • Use Amazon SageMaker with a suite of built-in algorithms to train and deploy machine learning models. AWS provides optimized versions of frameworks, such as TensorFlow, Chainer, Keras, and Theano. These frameworks include optimizations for high-performance training across Amazon EC2 instance families.

  • Use Amazon SageMaker Experiments to keep track of the models during testing.

  • Use Amazon SageMaker Autopilot to select algorithms automatically.

  • Discover pre-trained ML models on AWS Marketplace - Pre-trained ML models are ready-to-use models that can be quickly deployed on Amazon SageMaker. By pre-training the models for you, solutions in AWS Marketplace take care of the heavy lifting, helping your team deliver ML powered features faster and at a lower cost.

Documents

• Choose an Algorithm

• Manage Machine Learning with Amazon SageMakerExperiments

• Automate model development with Amazon SageMakerAutopilot

Blogs

• Optimizing costs for machine learning with Amazon SageMaker

• Amazon SageMaker Experiments – Organize, Track, and Compare Your Machine Learning Trainings

• Amazon SageMaker Autopilot – Automatically Create High-Quality Machine Learning Models with Full Control and Visibility

• Streamline modeling with Amazon SageMaker Studio and the Amazon Experiments SDK

Videos

• Accelerate Machine Learning Projects with Hundreds of Algorithms and Models in AWS Marketplace

• Organize, Track, and Evaluate ML Training Runs with Amazon SageMaker Experiments