Image
Settings used for training the model. For more information on the available settings please visit the official documentation: https://docs.microsoft.com/en-us/azure/machine-learning/how-to-auto-train-image-models.
Constructors
Properties
Settings for advanced scenarios.
Enable AMSGrad when optimizer is 'adam' or 'adamw'.
Settings for using Augmentations.
Frequency to store model checkpoints. Must be a positive integer.
The pretrained checkpoint model for incremental training.
The id of a previous run that has a pretrained checkpoint for incremental training.
Whether to use distributed training.
Enable early stopping logic during training.
Minimum number of epochs or validation evaluations to wait before primary metric improvement is tracked for early stopping. Must be a positive integer.
Minimum number of epochs or validation evaluations with no primary metric improvement before the run is stopped. Must be a positive integer.
Enable normalization when exporting ONNX model.
Frequency to evaluate validation dataset to get metric scores. Must be a positive integer.
Gradient accumulation means running a configured number of "GradAccumulationStep" steps without updating the model weights while accumulating the gradients of those steps, and then using the accumulated gradients to compute the weight updates. Must be a positive integer.
Number of layers to freeze for the model. Must be a positive integer. For instance, passing 2 as value for 'seresnext' means freezing layer0 and layer1. For a full list of models supported and details on layer freeze, please see: https://docs.microsoft.com/en-us/azure/machine-learning/how-to-auto-train-image-models.
Initial learning rate. Must be a float in the range 0, 1.
Type of learning rate scheduler. Must be 'warmup_cosine' or 'step'.
Number of training epochs. Must be a positive integer.
Number of data loader workers. Must be a non-negative integer.
Type of optimizer.
Random seed to be used when using deterministic training.
Value of gamma when learning rate scheduler is 'step'. Must be a float in the range 0, 1.
Value of step size when learning rate scheduler is 'step'. Must be a positive integer.
Training batch size. Must be a positive integer.
Image crop size that is input to the neural network for the training dataset. Must be a positive integer.
Validation batch size. Must be a positive integer.
Image crop size that is input to the neural network for the validation dataset. Must be a positive integer.
Image size to which to resize before cropping for validation dataset. Must be a positive integer.
Value of cosine cycle when learning rate scheduler is 'warmup_cosine'. Must be a float in the range 0, 1.
Value of warmup epochs when learning rate scheduler is 'warmup_cosine'. Must be a positive integer.
Value of weight decay when optimizer is 'sgd', 'adam', or 'adamw'. Must be a float in the range0, 1.
Weighted loss. The accepted values are 0 for no weighted loss. 1 for weighted loss with sqrt.(class_weights). 2 for weighted loss with class_weights. Must be 0 or 1 or 2.