Best practices by ML lifecycle

MLOPS01-BP01 Develop the right skills with accountability and empowerment

MLOPS01-BP02 Discuss and agree on the level of model explainability

MLOPS01-BP03 Monitor model adherence to business requirements

MLSEC01-BP01 Validate ML data permissions, privacy, software, and license terms

MLPERF01-BP01 Determine key performance indicators

MLCOST01-BP01 Define overall return on investment (ROI) and opportunity cost

MLCOST01-BP02 Use managed services to reduce total cost of ownership (TCO)

MLSUS01-BP01 Define the overall environmental impact or benefit

MLOPS02-BP01 Establish ML roles and responsibilities

MLOPS02-BP02 Prepare an ML profile template

MLOPS02-BP03 Establish model improvement strategies

MLOPS02-BP04 Establish a lineage tracker system

MLOPS02-BP05 Establish feedback loops across ML lifecycle phases

MLOPS02-BP06 Review fairness and explainability

MLSEC02-BP01 Design data encryption and obfuscation

MLREL01-BP01 Use APIs to abstract change from model consuming applications

MLREL01-BP02 Adopt a machine learning microservice strategy

MLPERF02-BP01 Define relevant evaluation metrics

MLPERF02-BP02 Use purpose-built AI and ML services and resources

MLCOST02-BP01 Identify if machine learning is the right solution

MLCOST02-BP02 Perform a tradeoff analysis between custom and pre-trained models

MLSUS02-BP01 Consider AI services and pre-trained models

MLSUS02-BP02 Select sustainable Regions

MLOPS03-BP01 Profile data to improve quality

MLOPS03-BP02 Create tracking and version control mechanisms

MLSEC03-BP01 Provide least privilege access

MLSEC03-BP02 Secure data and modeling environment

MLSEC03-BP03 Protect sensitive data privacy

MLSEC03-BP04 Enforce data lineage

MLSEC03-BP05 Keep only relevant data

MLREL02-BP01 Use a data catalog

MLREL02-BP02 Use a data pipeline

MLREL02-BP03 Automate managing data changes

MLPERF03-BP01 Use a modern data architecture

MLCOST03-BP01 Use managed data labeling

MLCOST03-BP02 Use no-code or low-code and code generation tools for interactive analysis

MLCOST03-BP03 Use managed data processing capabilities

MLCOST03-BP04 Enable feature reusability

MLSUS03-BP01 Minimize idle resources

MLSUS03-BP02 Implement data lifecycle policies aligned with your sustainability goals

MLSUS03-BP03 Adopt sustainable storage options

MLOPS04-BP01 Automate operations through MLOps and CI/CD

MLOPS04-BP02 Establish reliable packaging patterns to access approved public libraries

MLSEC04-BP01 Secure governed ML environment

MLSEC04-BP02 Secure inter-node cluster communications

MLSEC04-BP03 Protect against data poisoning threats

MLREL03-BP01 Enable CI/CD/CT automation with traceability

MLREL03-BP02 Verify feature consistency across training and inference

MLREL03-BP03 Validate models with relevant data

MLREL03-BP04 Establish data bias detection and mitigation

MLPERF04-BP01 Optimize training and inference instance types

MLPERF04-BP02 Explore alternatives for performance improvement

MLPERF04-BP03 Establish a model performance evaluation pipeline

MLPERF04-BP04 Establish feature statistics

MLPERF04-BP05 Perform a performance trade-off analysis

MLPERF04-BP06 Detect performance issues when using transfer learning

MLCOST04-BP01 Select optimal computing instance size

MLCOST04-BP02 Use managed build environments

MLCOST04-BP03 Select local training for small scale experiments

MLCOST04-BP04 Select an optimal ML framework

MLCOST04-BP05 Use automated machine learning

MLCOST04-BP06 Use managed training capabilities

MLCOST04-BP07 Use distributed training

MLCOST04-BP08 Stop resources when not in use

MLCOST04-BP09 Start training with small datasets

MLCOST04-BP10 Use warm start and checkpointing hyperparameter tuning

MLCOST04-BP11 Use hyperparameter optimization technologies

MLCOST04-BP12 Set up a budget and use resource tagging to track costs

MLCOST04-BP13 Enable data and compute proximity

MLCOST04-BP14 Select optimal algorithms

MLSUS04-BP01 Define sustainable performance criteria

MLSUS04-BP02 Select energy-efficient algorithms

MLSUS04-BP03 Archive or delete unnecessary training artifacts

MLSUS04-BP04 Use efficient model tuning methods

MLOPS05-BP01 Establish deployment environment metrics

MLSEC05-BP01 Protect against adversarial and malicious activities

MLREL04-BP01 Automate endpoint changes through a pipeline

MLREL04-BP02 Use an appropriate deployment and testing strategy

MLPERF05-BP01 Evaluate cloud versus edge options for machine learning deployment

MLPERF05-BP02 Choose an optimal deployment option in the cloud

MLCOST05-BP01 Use an appropriate deployment option

MLCOST05-BP02 Explore cost effective hardware options

MLCOST05-BP03 Right-size the model hosting instance fleet

MLSUS05-BP01 Align SLAs with sustainability goals

MLSUS05-BP02 Use efficient silicon

MLSUS05-BP03 Optimize models for inference

MLSUS05-BP04 Deploy multiple models behind a single endpoint

MLOPS06-BP01 Synchronize architecture and configuration, and check for skew across environments

MLOPS06-BP02 Enable model observability and tracking

MLSEC06-BP01 Restrict access to intended legitimate consumers

MLSEC06-BP02 Monitor human interactions with data for anomalous activity

MLREL05-BP01 Allow automatic scaling of the model endpoint

MLREL05-BP02 Create a recoverable endpoint with a managed version control strategy

MLPERF06-BP01 Include human-in-the-loop monitoring

MLPERF06-BP02 Evaluate model explainability

MLPERF06-BP03 Evaluate data drift

MLPERF06-BP04 Monitor, detect, and handle model performance degradation

MLPERF06-BP05 Establish an automated re-training framework

MLPERF06-BP06 Review for updated data and features for retraining

MLCOST06-BP01 Monitor usage and cost by ML activity

MLCOST06-BP02 Monitor return on investment for ML models

MLCOST06-BP03 Monitor endpoint usage and right-size the instance fleet

MLCOST06-BP04 Enable debugging and logging

MLSUS06-BP01 Measure material efficiency

MLSUS06-BP02 Retrain only when necessary