graph LR
tensorflow_addons_activations_sparsemax["tensorflow_addons.activations.sparsemax"]
tensorflow_addons_losses_giou_loss["tensorflow_addons.losses.giou_loss"]
tensorflow_addons_losses_triplet["tensorflow_addons.losses.triplet"]
tensorflow_addons_metrics_cohens_kappa["tensorflow_addons.metrics.cohens_kappa"]
tensorflow_addons_metrics_f_scores["tensorflow_addons.metrics.f_scores"]
tensorflow_addons_optimizers_adabelief["tensorflow_addons.optimizers.adabelief"]
tensorflow_addons_optimizers_lamb["tensorflow_addons.optimizers.lamb"]
The Fundamental ML Building Blocks subsystem provides specialized extensions to TensorFlow's core capabilities, organized into distinct functional areas: activations, loss functions, performance metrics, and optimization algorithms.
Computes the sparsemax activation function, ensuring sparse probability distributions. This is a specialized activation function extending TensorFlow's standard set.
Related Classes/Methods:
tensorflow_addons.activations.sparsemax:_swap_axistensorflow_addons.activations.sparsemax:_compute_2d_sparsemax
Provides the Generalized Intersection over Union (GIoU) loss for robust bounding box regression in object detection tasks. It addresses limitations of standard IoU loss by considering the enclosing box.
Related Classes/Methods:
Contains triplet loss functions for metric learning, focusing on learning embeddings where similar items are closer and dissimilar items are further apart by a margin.
Related Classes/Methods:
Computes Cohen's Kappa, a statistical measure of inter-rater agreement for categorical items, which is more robust than simple accuracy for imbalanced datasets.
Related Classes/Methods:
tensorflow_addons.metrics.cohens_kappa:_update_confusion_matrixtensorflow_addons.metrics.cohens_kappa:reset_state
Provides F-score metrics (e.g., F1-score) for evaluating classification models, particularly useful for imbalanced datasets where precision and recall are critical.
Related Classes/Methods:
Implements the AdaBelief optimizer, an adaptive learning rate algorithm that combines the benefits of Adam and RMSprop with improved stability.
Related Classes/Methods:
tensorflow_addons.optimizers.adabelief:_resource_apply_densetensorflow_addons.optimizers.adabelief:_resource_apply_sparse
Provides the LAMB (Layer-wise Adaptive Moments optimizer for Batching) optimizer, specifically optimized for training deep learning models with very large batch sizes.
Related Classes/Methods: