mojolearn.PageRank#

class mojolearn.PageRank(alpha=0.85, *, personalization=None, max_iter=100, tol=1e-06, nstart=None, weight=True, dangling=None)[source]#

Bases: _XNeighbors

PageRank scores of a directed graph by power iteration.

Follows networkx’s pagerank: a row-stochastic transition matrix, dangling nodes’ mass redistributed by dangling (or the personalization), and iteration until the summed absolute change is below n * tol. The graph is a weighted adjacency matrix with A[i, j] the weight of the edge i -> j, given as a dense square array, a scipy.sparse matrix (any format; its indices must be sorted within each row), or an (indptr, indices, n) CSR tuple in which every edge weighs 1. Vectors over nodes are length-n arrays (networkx uses dicts over nodes 0..n-1). Computation is float32.

Parameters:

  • alpha (float, default 0.85) – Damping factor.

  • personalization (array-like of shape (n,) or None, default None) – Teleport distribution, normalized to sum to one; None is uniform.

  • max_iter (int, default 100) – Maximum number of iterations. Not converging raises RuntimeError.

  • tol (float, default 1e-6) – Per-node convergence tolerance.

  • nstart (array-like of shape (n,) or None, default None) – Starting vector, normalized to sum to one; None is uniform.

  • weight (bool or None, default True) – Use the edge weights. False or None treats every edge as weight 1.

  • dangling (array-like of shape (n,) or None, default None) – Where the mass of nodes without out-edges goes, normalized to sum to one; None uses the personalization.

  • numeric_mode ({‘identical’, ‘deterministic’, ‘fast’} or None, default None) – Numeric mode for this estimator ('fast' is Apple only). None uses the process default.

  • The ``personalization``, ``nstart`` and ``dangling`` vectors must have

  • length n, be non-negative and not all zero (``ValueError`` otherwise).

Variables:

  • pagerank_ (Array of shape (n,), float32) – PageRank score of each node.

  • n_iter_ (int) – Number of iterations run.

Examples

>>> import numpy as np, mojolearn
>>> A = (np.random.default_rng(0).random((500, 500)) < 0.01).astype(np.float32)
>>> scores = mojolearn.PageRank(alpha=0.85).fit(A).pagerank_
__init__(alpha=0.85, *, personalization=None, max_iter=100, tol=1e-06, nstart=None, weight=True, dangling=None)[source]#

Methods

fit(A[, y])

Compute the PageRank scores of the graph.

get_params([deep])

numeric_mode_used()

The numeric mode this estimator runs in.

set_params(**params)

Set constructor parameters.

vendor_used()

The GPU backend this estimator runs on.
fit(A, y=None)[source]#

Compute the PageRank scores of the graph.

Parameters:

  • A (array-like of shape (n, n), scipy.sparse matrix, or tuple) – Weighted adjacency matrix (A[i, j] is the weight of i -> j): a dense square array, a scipy.sparse matrix, or an (indptr, indices, n) CSR tuple.

  • y (None) – Ignored.

Returns:

self (PageRank) – The fitted estimator, with pagerank_ and n_iter_.

Raises:

RuntimeError – If the iteration does not converge within max_iter steps.