Watch keynotes, product sessions, workshops, and more from Google I/O See playlist

Module: nsl.configs

Configuration classes and APIs for Neural Structured Learning.

Classes

class AdvNeighborConfig: Contains configuration for generating adversarial neighbors.

class AdvRegConfig: Contains configuration for adversarial regularization.

class AdvTargetConfig: Contains configuration for selecting targets to be attacked.

class AdvTargetType: Types of adversarial targeting.

class DecayConfig: Contains configuration for decaying a value during training.

class DecayType: Types of decay.

class DistanceConfig: Contains configuration for computing distances between tensors.

class DistanceType: Types of distance.

class GraphBuilderConfig: Encapsulates configuration parameters for building a graph.

class GraphNeighborConfig: Specifies neighbor attributes for graph regularization.

class GraphRegConfig: Contains the configuration for graph regularization.

class IntegrationConfig: Contains configuration for computing multimodal integration.

class IntegrationType: Types of integration for multimodal fusion.

class NormType: Types of norms.

class TransformType: Types of nonlinear functions to be applied .

class VirtualAdvConfig: Contains configuration for virtual adversarial training.

Functions

make_adv_reg_config(...): Creates an nsl.configs.AdvRegConfig object.

make_graph_reg_config(...): Creates an nsl.configs.GraphRegConfig object.

DEFAULT_ADVERSARIAL_PARAMS

{
 'adv_grad_norm': <NormType.L2: 'l2'>,
 'adv_step_size': 0.001,
 'clip_value_max': None,
 'clip_value_min': None,
 'feature_mask': None,
 'pgd_epsilon': None,
 'pgd_iterations': 1
}

DEFAULT_DISTANCE_PARAMS

{
 'distance_type': <DistanceType.L2: 'l2'>,
 'reduction': 'weighted_sum_by_nonzero_weights',
 'sum_over_axis': None,
 'transform_fn': <TransformType.NONE: 'none'>
}