WEFE API

This reference details all the utilities as well as the metrics and mitigation methods implemented so far in WEFE.

WordEmbeddingModel

wefe.word_embedding_model.WordEmbeddingModel(wv)

A wrapper for Word Embedding pre-trained models.

Query

wefe.query.Query(target_sets, attribute_sets)

A container for attribute and target word sets.

Metrics

This list contains the metrics implemented in WEFE.

`wefe.metrics.WEAT`()	Word Embedding Association Test (WEAT).
`wefe.metrics.RND`()	Relative Norm Distance (RND).
`wefe.metrics.RNSB`()	Relative Relative Negative Sentiment Bias (RNSB).
`wefe.metrics.MAC`()	Mean Average Cosine Similarity (MAC).
`wefe.metrics.ECT`()	Embedding Coherence Test (ECT).
`wefe.metrics.RIPA`()	Relational Inner Product Association Test (RIPA).

Debias

This list contains the debiasing methods implemented so far in WEFE.

`wefe.debias.hard_debias.HardDebias`([...])	Hard Debias debiasing method.
`wefe.debias.multiclass_hard_debias.MulticlassHardDebias`([...])	Generalized version of Hard Debias that enables multiclass debiasing.
`wefe.debias.repulsion_attraction_neutralization.RepulsionAttractionNeutralization`([...])	Repulsion Attraction Neutralization method.
`wefe.debias.double_hard_debias.DoubleHardDebias`([...])	Double Hard Debias Method.
`wefe.debias.half_sibling_regression.HalfSiblingRegression`([...])	Half Sibling Debias method.

Datasets

The following functions allow you to load sets of words used in previous studies.

`wefe.datasets.load_bingliu`()	Load the Bing-Liu sentiment lexicon.
`wefe.datasets.fetch_debias_multiclass`()	Fetch the word sets used in the paper Black Is To Criminals as Caucasian
`wefe.datasets.fetch_debiaswe`()	Fetch the word sets used in the paper Man is to Computer Programmer as Woman is to Homemaker? from the source.
`wefe.datasets.fetch_eds`([occupations_year, ...])	Fetch the sets of words used in the experiments of the _Word Embeddings
`wefe.datasets.load_weat`()	Load the word sets used in the experiments of the

Preprocessing

The following functions allow transforming sets of words and queries to embeddings. The documentation of the functions in this section are intended as a guide for WEFE developers.

`wefe.preprocessing.preprocess_word`(word[, ...])	pre-processes a word before it is searched in the model's vocabulary.
`wefe.preprocessing.get_embeddings_from_set`(...)	Transform a sequence of words into dictionary that maps word - word embedding.
`wefe.preprocessing.get_embeddings_from_tuples`(...)	Given a sequence of word sets, obtain their corresponding embeddings.
`wefe.preprocessing.get_embeddings_from_query`(...)	Obtain the word vectors associated with the provided Query.

Utils

Collection of assorted utils.

`wefe.utils.load_test_model`()	Load a Word2vec subset to test metrics and debias methods.
`wefe.utils.generate_subqueries_from_queries_list`(...)	Generate a list of subqueries from queries.
`wefe.utils.run_queries`(metric, queries, models)	Run several queries over a several word embedding models using a specific metic.
`wefe.utils.plot_queries_results`(results[, by])	Plot the results obtained by a run_queries execution.
`wefe.utils.create_ranking`(results_dataframes)	Create a ranking form the aggregated scores of the provided dataframes.
`wefe.utils.plot_ranking`(ranking[, ...])
`wefe.utils.calculate_ranking_correlations`(...)	Calculate the correlation between the calculated rankings.
`wefe.utils.plot_ranking_correlations`(...[, ...])
`wefe.utils.flair_to_gensim`(flair_embedding)