pe.histogram package

Subpackages

Submodules

pe.histogram.histogram module

class pe.histogram.histogram.Histogram[source]

Bases: ABC

The abstract class for computing the histogram over synthetic samples. The histogram values indicate how good each synthetic sample is in terms their closeness to the private data.

abstract compute_histogram(priv_data, syn_data)[source]

Compute the histogram over the synthetic data using the private data.

Parameters:

pe.histogram.nearest_neighbors module

class pe.histogram.nearest_neighbors.NearestNeighbors(embedding, mode, lookahead_degree, lookahead_log_folder=None, voting_details_log_folder=None, api=None, num_nearest_neighbors=1, backend='auto')[source]

Bases: Histogram

Compute the nearest neighbors histogram. Each private sample will vote for their closest num_nearest_neighbors synthetic samples to construct the histogram. The l2 norm of the votes from each private sample is normalized to 1.

__init__(embedding, mode, lookahead_degree, lookahead_log_folder=None, voting_details_log_folder=None, api=None, num_nearest_neighbors=1, backend='auto')[source]

Constructor.

Parameters:

  • embedding (pe.embedding.embedding.Embedding) – The pe.embedding.embedding.Embedding object to compute the embedding of samples

  • mode (str) – The distance metric to use for finding the nearest neighbors. It should be one of the following: “l2” (l2 distance), “cos_sim” (cosine similarity), “ip” (inner product). Not all backends support all modes

  • lookahead_degree (int) – The degree of lookahead to compute the embedding of synthetic samples. If it is 0, the original embedding is used. If it is greater than 0, the embedding of the synthetic samples is computed by averaging the embeddings of the synthetic samples generated by the variation API for lookahead_degree times

  • lookahead_log_folder (str, optional) – The folder to save the logs of the lookahead. If it is None, the logs are not saved. Defaults to None

  • voting_details_log_folder (str, optional) – The folder to save the logs of the voting details. If it is None, the logs are not saved. Defaults to None

  • api (pe.api.api.API, optional) – The API to generate synthetic samples. It should be provided when lookahead_degree is greater than 0. Defaults to None

  • num_nearest_neighbors (int, optional) – The number of nearest neighbors to consider for each private sample, defaults to 1

  • backend (str, optional) – The backend to use for finding the nearest neighbors. It should be one of the following: “faiss” (FAISS), “sklearn” (scikit-learn), “auto” (using FAISS if available, otherwise scikit-learn). Defaults to “auto”. FAISS supports GPU and is much faster when the number of synthetic samples and/or private samples is large. It requires the installation of faiss-gpu or faiss-cpu package. See https://faiss.ai/

Raises:

  • ValueError – If the api is not provided when lookahead_degree is greater than 0

  • ValueError – If the backend is unknown

_compute_lookahead_embedding(syn_data)[source]

Compute the embedding of synthetic samples with lookahead.

Parameters:

syn_data (pe.data.data.Data) – The synthetic data

Returns:

The synthetic data with the computed embedding in the column pe.constant.data.LOOKAHEAD_EMBEDDING_COLUMN_NAME

Return type:

pe.data.data.Data

_log_lookahead(syn_data, lookahead_id)[source]

Log the lookahead data.

Parameters:

  • syn_data (pe.data.data.Data) – The lookahead data

  • lookahead_id (int) – The ID of the lookahead

_log_voting_details(priv_data, syn_data, ids)[source]

Log the voting details.

Parameters:

  • priv_data (pe.data.data.Data) – The private data

  • syn_data (pe.data.data.Data) – The synthetic data

  • ids (np.ndarray) – The IDs of the nearest neighbors for each private sample

compute_histogram(priv_data, syn_data)[source]

Compute the nearest neighbors histogram.

Parameters:
Returns:

The private data, possibly with the additional embedding column, and the synthetic data, with the computed histogram in the column pe.constant.data.CLEAN_HISTOGRAM_COLUMN_NAME and possibly with the additional embedding column

Return type:

tuple[pe.data.data.Data, pe.data.data.Data]