Pipeline component for image classification.

Version: 0.0.21

View in Studio: https://ml.azure.com/registries/azureml/components/image_classification_pipeline/version/0.0.21

------------------- Computes -------------------

Name	Description	Type	Default	Optional	Enum
compute_model_import	Compute to be used for framework_selector eg. provide 'cpu-cluster' if your compute is named 'cpu-cluster'.	string		False
compute_finetune	Compute to be used for running the selected framework eg. provide 'gpu-cluster' if your compute is named 'gpu-cluster'.	string		False
instance_count	Number of nodes to be used for finetuning (used for distributed training)	integer	1	True
process_count_per_instance	Number of gpus to be used per node for finetuning, should be equal to number of gpu per node in the compute SKU used for finetune.	integer	1	True

------------------- Model Framework Selector -------------------

Name	Description	Type	Default	Optional	Enum
model_name	Name of the model. Based on this model name, a framework will be selected (Hugging Face, MM Detection).	string		True
download_from_source	Download model directly from HuggingFace instead of system registry	boolean	False	True

------------------- Data Inputs ------------------

Name	Description	Type	Default	Optional	Enum
training_data	Path to MLTable for training data.	mltable		False
validation_data	Path to MLTable for validation data.	mltable		True

------------------- Classification Type ------------------

Name	Description	Type	Default	Optional	Enum
task_type	Whether a single image can have multiple labels.	string			['image-classification', 'image-classification-multilabel']
primary_metric	Primary metric for the task	string		True	['accuracy', 'iou']
ams_gradient	Enable ams_gradient when optimizer is adam or adamw.	boolean		True
beta1	Value of beta1 when optimizer is adam or adamw. Must be a float in the range [0, 1].	number		True
beta2	Value of beta2 when optimizer is adam or adamw. Must be a float in the range [0, 1].	number		True
checkpoint_frequency	Frequency to store model checkpoints. Must be a positive integer.	integer		True
checkpoint_run_id	The run ID of the experiment that has a pretrained checkpoint for incremental training.	string		True
early_stopping	Enable early stopping logic during training.	boolean		True
early_stopping_patience	Minimum number of epochs or validation evaluations with no primary metric improvement before the run is stopped. Must be a positive integer.	integer		True
early_stopping_delay	Minimum number of epochs or validation evaluations to wait before primary metric improvement is tracked for early stopping. Must be a positive integer.	integer		True
evaluation_frequency	Frequency to evaluate validation dataset to get metric scores. Must be a positive integer.	integer		True
gradient_accumulation_step	Number of forward passes 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.	integer		True
layers_to_freeze	How many layers to freeze for your model. For instance, passing 2 as value for seresnext means freezing layer0 and layer1 referring to the below supported model layer info. Must be a positive integer.	integer		True
learning_rate	Initial learning rate.	number		True
learning_rate_scheduler	Type of learning rate scheduler. Must be warmup_cosine or step.	string		True	['warmup_cosine', 'step']
momentum	Value of momentum when optimizer is sgd. Must be a float in the range [0, 1].	number		True
nesterov	Enable nesterov when optimizer is sgd.	boolean		True
number_of_epochs	Number of training epochs	integer		True
number_of_workers	Number of subprocesses to use for data loading (PyTorch only). 0 means that the data will be loaded in the main process.	integer		True
optimizer	Type of optimizer	string		True	['sgd', 'adam', 'adamw']
random_seed	Random seed that will be set at the beginning of training.	integer		True
step_lr_gamma	Value of gamma when learning rate scheduler is step. Please check for https://learn.microsoft.com/azure/machine-learning/reference-automl-images-hyperparameters more information.	number		True
step_lr_step_size	Value of step size when learning rate scheduler is step. Please check for https://learn.microsoft.com/azure/machine-learning/reference-automl-images-hyperparameters more information.	integer		True
training_batch_size	Training batch size.	integer		True
training_crop_size	Image crop size that's input to your neural network for training dataset. Notes - seresnext doesn't take an arbitrary size. ViT-variants should have the same validation_crop_size and training_crop_size.	integer		True
validation_batch_size	Validation batch size.	integer		True
validation_crop_size	Image crop size that's input to your neural network for validation dataset. Note - seresnext doesn't take an arbitrary size. ViT-variants should have the same validation_crop_size and training_crop_size.	integer		True
validation_resize_size	Image size to which to resize before cropping for validation dataset. Note - seresnext doesn't take an arbitrary size.	integer		True
warmup_cosine_lr_cycles	Value of cosine cycle when learning rate scheduler is warmup_cosine. Please check for https://learn.microsoft.com/azure/machine-learning/reference-automl-images-hyperparameters more information.	number		True
warmup_cosine_lr_warmup_epochs	Value of warmup epochs when learning rate scheduler is warmup_cosine. Please check for https://learn.microsoft.com/azure/machine-learning/reference-automl-images-hyperparameters more information.	integer		True
weight_decay	Value of weight decay used by the optimizer.	number		True
weighted_loss	Value of weighted loss.	integer		True

Name	Description	Type
pytorch_model_folder	The trained pytorch model.	custom_model
mlflow_model_folder	The trained MLFlow model.	mlflow_model