Sparse#
2026-09-18
3 min read time
This page provides pylibraft class references for the publicly-exposed elements of the pylibraft.sparse.linalg package.
- linalg.lanczos.eigsh(*args, handle=None, **kwargs)#
eigsh(A, k=6, which=’LM’, v0=None, ncv=None, maxiter=None, tol=0, seed=None, handle=None)
Find
keigenvalues and eigenvectors of the real symmetric square matrix or complex Hermitian matrixA.Solves
Ax = wx, the standard eigenvalue problem forweigenvalues with corresponding eigenvectorsx.- Parameters:
a (spmatrix) – A symmetric square sparse CSR matrix with dimension
(n, n).amust be of typecupyx.scipy.sparse._csr.csr_matrixk (int) – The number of eigenvalues and eigenvectors to compute. Must be
1 <= k < n.which (str) – ‘LM’ or ‘LA’ or ‘SA’. ‘LM’: finds
klargest (in magnitude) eigenvalues. ‘LA’: findsklargest (algebraic) eigenvalues. ‘SA’: findsksmallest (algebraic) eigenvalues. ‘SM’: findsksmallest (in magnitude) eigenvalues.v0 (ndarray) – Starting vector for iteration. If
None, a random unit vector is used.ncv (int) – The number of Lanczos vectors generated. Must be
k + 1 < ncv < n. IfNone, default value is used.maxiter (int) – Maximum number of Lanczos update iterations. If
None, default value is used.tol (float) – Tolerance for residuals
||Ax - wx||. If0, machine precision is used.
- Returns:
It returns
wandxwherewis eigenvalues andxis eigenvectors.- Return type:
tuple
See also
scipy.sparse.linalg.eigsh()cupyx.scipy.sparse.linalg.eigsh()Note
This function uses the thick-restart Lanczos methods (https://sdm.lbl.gov/~kewu/ps/trlan.html).
- linalg.svds.svds(*args, handle=None, **kwargs)#
svds(A, k=6, n_oversamples=10, n_power_iters=2, seed=None, return_singular_vectors=True, handle=None)
Compute the largest
ksingular values and corresponding singular vectors of a sparse matrix using randomized SVD.Computes the truncated SVD:
A ~ U @ diag(S) @ Vt.- Parameters:
A (cupyx.scipy.sparse.csr_matrix) – Sparse CSR matrix of shape
(m, n). Must be of typecupyx.scipy.sparse.csr_matrix.k (int) – Number of singular values and vectors to compute. Must be
1 <= k < min(m, n). Default 6.n_oversamples (int) – Number of extra random vectors for better approximation. Total subspace dimension is
k + n_oversamples. Default 10.n_power_iters (int) – Number of power iteration passes. More iterations improve accuracy for matrices with slowly decaying singular values. Default 2.
seed (int or None) – Random seed for reproducibility. If
None, a non-deterministic seed is used.return_singular_vectors (bool or {"u", "vh"}) –
Controls which singular vectors are returned (matches
scipy.sparse.linalg.svds()).True(default): return(U, S, Vt).False: skip both vector matrices, return(None, S, None)."u": skipVt, return(U, S, None)."vh": skipU, return(None, S, Vt).
Skipping a side avoids the corresponding output buffer and final matrix multiplication.
handle – RAFT resource handle. If
None, a default is created.
- Returns:
(U, S, Vt)whereUis left singular vectors(m, k),Sis singular values(k,)in descending order, andVtis right singular vectors(k, n).Uand/orVtmay beNonedepending onreturn_singular_vectors.- Return type:
tuple
See also
scipy.sparse.linalg.svds()Note
This function uses randomized SVD (Halko et al. 2009) with CholeskyQR2 orthogonalization (Tomas et al. 2024) for efficient GPU execution.