networks.embed.py

class apparent.networks.embed.NetworkEmbedder(pairwise_distances=None, method=<class 'sklearn.manifold._t_sne.TSNE'>, **kwargs)

A class for embedding networks into a lower-dimensional space using pairwise distances or raw data.

This class provides functionality to embed networks into a lower-dimensional space for visualization and analysis. It supports both precomputed distance matrices and raw data, using t-SNE as the default embedding method.

Parameters:

• pairwise_distances (np.ndarray, optional) – A pairwise distance matrix (n x n) between networks. Used for precomputed embedding methods.

• method (object, optional) – An embedding method class. Default is TSNE from scikit-learn.

• **kwargs (dict) – Additional arguments passed to the embedding method.

D

A pairwise distance matrix (n x n) between networks.

Type:

np.ndarray or None

method

An embedding method class (default: TSNE).

Type:

object

scaler

Scaler for standardizing the embedding output.

Type:

StandardScaler

kwargs

Additional arguments for the embedding method.

Type:

dict

Examples

>>> import numpy as np
>>> from apparent.networks import NetworkEmbedder
>>>
>>> # Create sample distance matrix
>>> distances = np.random.rand(10, 10)
>>> distances = (distances + distances.T) / 2  # Make symmetric
>>> np.fill_diagonal(distances, 0)  # Zero diagonal
>>>
>>> # Create embedding
>>> embedder = NetworkEmbedder(pairwise_distances=distances)
>>> embedding = embedder.embed()
>>> print(f"Embedding shape: {embedding.shape}")

__init__(pairwise_distances=None, method=<class 'sklearn.manifold._t_sne.TSNE'>, **kwargs)

Initialize the NetworkEmbedder.

Parameters:

• pairwise_distances (np.ndarray, optional) – Pairwise distance matrix for precomputed embedding methods. Should be a symmetric matrix with zeros on the diagonal.

• method (object, optional) – The embedding method class to use. Default is TSNE from scikit-learn. Must implement fit_transform method.

• **kwargs (dict) – Additional arguments for the embedding method. Common parameters for t-SNE include ‘n_components’, ‘perplexity’, ‘learning_rate’.

Examples

>>> embedder = NetworkEmbedder()  # Default t-SNE
>>> embedder = NetworkEmbedder(method=TSNE, perplexity=30.0)
>>> embedder = NetworkEmbedder(pairwise_distances=distances)

embed(data=None)

Perform the embedding based on the selected method.

This method creates a low-dimensional embedding of the networks using either precomputed pairwise distances or raw data. The embedding is standardized using StandardScaler before being returned.

Parameters:

data (np.ndarray, optional) – Raw input data for embedding (used if pairwise_distances is None). Should be a 2D array with shape (n_samples, n_features).

Returns:

embedding – Low-dimensional embedding of the networks, typically 2D for visualization. The embedding is standardized to have zero mean and unit variance.

Return type:

np.ndarray

Raises:

ValueError – If neither pairwise distances nor data are provided.

Examples

>>> embedder = NetworkEmbedder(pairwise_distances=distances)
>>> embedding = embedder.embed()
>>> print(f"Embedding shape: {embedding.shape}")  # (n_networks, 2)
>>>
>>> # Using raw data instead
>>> embedder = NetworkEmbedder()
>>> embedding = embedder.embed(data=feature_matrix)