langspace.probe package

Subpackages

• langspace.probe.arithmetic package
  • Module contents
    • ArithmeticProbe
      • ArithmeticProbe.model
      • ArithmeticProbe.data
      • ArithmeticProbe.ops
      • ArithmeticProbe.annotations
      • ArithmeticProbe.arithmetic()
      • ArithmeticProbe.report()
• langspace.probe.cluster_vis package
  • Submodules
  • langspace.probe.cluster_vis.methods module
    • ClusterVisualizationMethod
      • ClusterVisualizationMethod.PCA
      • ClusterVisualizationMethod.TSNE
      • ClusterVisualizationMethod.UMAP
  • Module contents
    • ClusterVisualizationProbe
      • ClusterVisualizationProbe.model
      • ClusterVisualizationProbe.data
      • ClusterVisualizationProbe.sample_size
      • ClusterVisualizationProbe.target_roles
      • ClusterVisualizationProbe.method
      • ClusterVisualizationProbe.cluster_annot
      • ClusterVisualizationProbe.batch_size
      • ClusterVisualizationProbe.annotations
      • ClusterVisualizationProbe.plot_label_map
      • ClusterVisualizationProbe.report()
      • ClusterVisualizationProbe.role_content_viz()
      • ClusterVisualizationProbe.structure_viz()
• langspace.probe.defmod package
  • Module contents
• langspace.probe.disentanglement package
  • Module contents
    • DisentanglementProbe
      • DisentanglementProbe.beta_vae_metric()
      • DisentanglementProbe.categorical_crossentropy_loss()
      • DisentanglementProbe.disentanglement_completeness_informativeness()
      • DisentanglementProbe.entropy()
      • DisentanglementProbe.factor_vae_metric()
      • DisentanglementProbe.group_sampling()
      • DisentanglementProbe.modularity_explicitness()
      • DisentanglementProbe.mutual_information_estimation()
      • DisentanglementProbe.mutual_information_gap()
      • DisentanglementProbe.report()
      • DisentanglementProbe.separated_attribute_predictability()
      • DisentanglementProbe.stratified_sampling()
    • GenerativeDataset
      • GenerativeDataset.generative_factors
      • GenerativeDataset.value_space
      • GenerativeDataset.sample_space
      • GenerativeDataset.representation_space
      • GenerativeDataset.get_representation_space()
    • SRLFactorDataset
      • SRLFactorDataset.generative_factors
      • SRLFactorDataset.value_space
      • SRLFactorDataset.sample_space
      • SRLFactorDataset.structure
• langspace.probe.interpolation package
  • Module contents
    • InterpolationProbe
      • InterpolationProbe.model
      • InterpolationProbe.data
      • InterpolationProbe.eval
      • InterpolationProbe.annotations
      • InterpolationProbe.report()
• langspace.probe.lingprop package
  • Module contents
    • LinguisticPropertiesProbe
      • LinguisticPropertiesProbe.report()
• langspace.probe.sts package
  • Module contents
    • STSProbe
      • STSProbe.report()
• langspace.probe.traversal package
  • Module contents
    • TraversalProbe
      • TraversalProbe.dims
      • TraversalProbe.model
      • TraversalProbe.sample_size
      • TraversalProbe.annotations
      • TraversalProbe.report()

Submodules

langspace.probe.base module

class langspace.probe.base.LatentSpaceProbe(model: LangVAE, data: Iterable[Sentence], sample_size: int, **kwargs)

Bases: ABC

Abstract base class for probing the latent space of a language VAE.

batched_encoding(data: Iterable[Sentence], annotations: Dict[str, List[str]] = None, batch_size: int = 100) → Tensor

Encodes the sentences

Parameters:

• data (Iterable[Sentence]) – sentences

• annotations (List[str]) – optional annotations to be used, if available in the data, and their respective

• values. (possible) –

• batch_size (int) – number of sentences to be processed simultaneously

Returns:

Latent representation

Return type:

Tensor

decoding(prior: Tensor, cvars_emb: List[Tensor] = None) → List[str]

args: sent_num by latent_dim return: sentence list

encoding(data: Iterable[Sentence], annotations: Dict[str, List[str]] = None) → Tuple[Tensor, Tensor, Tensor, List[Tensor]]

Encode the input data and return the mean, standard deviation, and latent representation.

Parameters:

data (Iterable[Union[str, Sentence]]) – The input data to encode.

Returns:

A tuple containing the mean, standard deviation, latent representation and conditional variable embeddings, as tensors.

Return type:

Tuple[Tensor, Tensor, Tensor, Tensor]

get_tokenized_data_seed(data: Iterable[Sentence], annotations: Dict[str, List[str]] = None) → TokenizedDataSet

abstract report() → DataFrame

Generate a report from the probe.

Returns:

The generated report.

Return type:

DataFrame

Module contents