Version: 0.0.19

Preview

View in Studio: https://ml.azure.com/registries/azureml/components/rai_vision_insights/version/0.0.19

Name	Description	Type	Default	Optional	Enum
task_type	The type of task to perform	string			['image_classification', 'multilabel_image_classification', 'object_detection']
model_input	The model name to be used for computing insights	mlflow_model		False
model_info	The model name to be used for computing insights	string		False
test_dataset	The test dataset to be used for computing insights	mltable
target_column_name	The target column name	string
maximum_rows_for_test_dataset	The maximum number of rows to use from the test dataset	integer	5000
classes	The list of class names for the target column	string	[]
categorical_metadata_features	The list of categorical metadata feature names	string	[]
dropped_metadata_features	The list of dropped metadata feature names	string	[]
precompute_explanation	Whether to precompute explanations	boolean	True
enable_error_analysis	Whether to enable computation of error analysis	boolean	True
use_model_dependency	Whether to install the MLFlow model's dependencies in the RAI environment	boolean	False
use_conda	Whether to use conda to install dependencies	boolean	False
model_type	The type of MLFlow model to deserialize	string			['pyfunc', 'fastai', 'pytorch']

guided_gradcam doesn't work with transformer vision models and shap isn't supported for automl images models for more details on XAI parameters, refer to following link https://learn.microsoft.com/en-us/azure/machine-learning/how-to-auto-train-image-models?tabs=cligenerate-explanations-for-predictions

Name	Description	Type	Default	Optional	Enum
xai_algorithm	The explanation algorithm to use for AutoML vision models, always set to shap for others	string	guided_backprop	True	['guided_backprop', 'guided_gradcam', 'integrated_gradients', 'xrai', 'shap']
n_steps	The number of steps for the integrated gradients and XRAI algorithms	integer		True
xrai_fast	Whether to use the faster version of the XRAI algorithm	boolean		True
approximation_method	The approximation method to use for the integrated gradients algorithm	string		True	['gausslegendre', 'riemann_middle']
confidence_score_threshold_multilabel	The confidence score threshold for multilabel classification explanations, above which the labels are selected for generating explanations	number		True
image_width_in_inches	The width to resize the image to in inches	number		True
max_evals	The maximum number of evaluations to run in shap's hierarchical image explainer.	integer		True
num_masks	The number of masks to use for the DRISE image explainer for object detection.	integer		True
mask_res	The resolution of the masks to use for the DRISE image explainer for object detection.	integer		True
dataset_type	The type of image dataset to use, whether the images are on private azure blob storage or public urls	string	public		['private', 'public']

Name	Description	Type
dashboard	Path to which RAIVisionInsights is serialized to for connecting to compute instance	path
ux_json	Json file to which UX is serialized to for viewing in static AzureML Studio UI	path

azureml://registries/azureml/environments/responsibleai-vision/versions/7