languagechange.models.change package

Submodules

languagechange.models.change.metrics module

class languagechange.models.change.metrics.ChangeModel

Bases: object

class languagechange.models.change.metrics.BinaryChange

Bases: ChangeModel

predict()

class languagechange.models.change.metrics.GradedChange

Bases: ChangeModel

compute_scores()

class languagechange.models.change.metrics.Threshold

Bases: BinaryChange

set_threshold(threshold)

class languagechange.models.change.metrics.AutomaticThrehold

Bases: Threshold

compute_threshold(scores, func=<function AutomaticThrehold.<lambda>>)

class languagechange.models.change.metrics.OptimalThrehold

Bases: Threshold

compute_threshold(scores, vrange=numpy.arange, evaluator=None)

class languagechange.models.change.metrics.APD

Bases: GradedChange

compute_scores(embeddings1, embeddings2, metric='cosine')

class languagechange.models.change.metrics.PRT

Bases: GradedChange

compute_scores(embeddings1, embeddings2, metric='cosine')

class languagechange.models.change.metrics.PJSD

Bases: GradedChange

compute_scores(embeddings1, embeddings2, clustering_algorithm, metric='cosine')

languagechange.models.change.timeseries module

languagechange.models.change.timeseries.ma(ts, k)

Computes the moving average of a timeseries. :param ts: a timeseries. :type ts: np.array :param k: the window (k timesteps to the left and k to the right) :type k: int

Returns:

the moving average of the timeseries (not including endpoints)

class languagechange.models.change.timeseries.TimeSeries(embs=None, series=None, change_metric=None, timeseries_type=None, k=1, time_labels=None, clustering_algorithm=None, distance_metric='cosine')

Bases: object

Parameters:

• embs (List[numpy.array])

• series (numpy.array)

• timeseries_type (str)

• time_labels (numpy.array | List)

compute_from_embeddings(embs, change_metric, timeseries_type, k=1, time_labels=None, clustering_algorithm=None, distance_metric='cosine')

Parameters:

• embs ([np.array]) – a list of embeddings, each element of the list contains embeddings from one time period.

• change_metric (str|object) – the metric to use when comparing embeddings from different time periods (should be one of the classes in languagechange.models.change.metrics).

• timeseries_type (str) – the kind of timeseries to construct. One of [‘compare_to_first’, ‘compare_to_last’, ‘consecutive’, ‘moving_average’].

• time_labels (np.array|list) – labels for the x axis of the timeseries.

• clustering_algorithm – the clustering algorithm if using PJSD as the change metric. E.g. one of the algorithms in scikit-learn, or languagechange.

• distance_metric (str) – the distance metric to use when computing change scores.

Returns:

the final timeseries. ts (np.array): the time values/labels for each value in the final timeseries.

Return type:

series (np.array)

languagechange.models.change.widid module

class languagechange.models.change.widid.WiDiD(algorithm=<class 'languagechange.models.meaning.clustering.APosterioriaffinityPropagation'>, metric='cosine', **args)

Bases: object

A class that implements WiDiD (https://github.com/FrancescoPeriti/WiDiD).

compute_scores(embs_list, timeseries_type='consecutive', k=1, change_metric='apd', time_labels=None)

Performs a-posteriori affinity propagation (APP) clustering and computes the semantic change as the APD (or another metric) between the prototype embeddings in clusters of different time periods.

Parameters:

• embs_list ([np.array]) – a list of embeddings for a target word, where each element is embeddings of one time period.

• timeseries_type (str) – the type of timeseries (see usage in languagechange.models.change.timeseries).

• k (int) – the window size, if moving average (see usage in languagechange.models.change.timeseries).

• change_metric (str) – the change metric (e.g. ‘apd’) to use (see usage in languagechange.models.change.timeseries).

• change_metric – the change metric (e.g. ‘apd’) to use (see usage in languagechange.models.change.timeseries).

• time_labels (np.array|list) – labels for the x axis of the timeseries (see usage in languagechange.models.change.timeseries).

Returns:

the labels for each embedding in each time period. prot_embs ([np.array]): a list of matrices encoding the prototype (average) embedding of each cluster in each time period. change_scores (TimeSeries): a timeseries (languagechange.models.change.timeseries.TimeSeries) containing the degree of change between the embeddings in different time periods.

Return type:

labels ([np.array])

Module contents