Built-in Rules Provided by Amazon SageMaker Debugger - Amazon SageMaker

Built-in Rules Provided by Amazon SageMaker Debugger

Use the built-in rules provided by Amazon SageMaker Debugger to analyze tensors emitted during the training of machine learning models. These rules monitor various common conditions that are critical for the success of a training job. There are four scopes of validity for the built-in rules.

Scopes of Validity for Build-in Rules
Scope of Validity Built-in Rules
Deep learning frameworks (TensorFlow, Apache MXNet, and PyTorch)
  • DeadRelu

  • ExplodingTensor

  • PoorWeightInitialization

  • SaturatedActivation

  • VanishingGradient

  • WeightUpdateRatio

Deep learning frameworks (TensorFlow, MXNet, and PyTorch) and the XGBoost algorithm
  • AllZero

  • ClassImbalance

  • Confusion

  • LossNotDecreasing

  • Overfit

  • Overtraining

  • SimilarAcrossRuns

  • TensorVariance

  • UnchangedTensor

Deep learning applications
  • CheckInputImages

  • NLPSequenceRatio

XGBoost algorithm
  • TreeDepth