parsimony.algorithms.nipals.FastSparseSVD

Class FastSparseSVD

object --+ | bases.BaseAlgorithm --+ | bases.ImplicitAlgorithm --+ | FastSparseSVD

Instance Methods

run(self, X, max_iter=100, start_vector=None)
A kernel SVD implementation for sparse CSR matrices.

source code

Inherited from bases.BaseAlgorithm: get_params, set_params

Inherited from object: __delattr__, __format__, __getattribute__, __hash__, __init__, __new__, __reduce__, __reduce_ex__, __repr__, __setattr__, __sizeof__, __str__, __subclasshook__

Class Variables

[hide private]

__abstractmethods__ = frozenset([])

_abc_negative_cache_version = 14

Inherited from bases.ImplicitAlgorithm: __metaclass__

Inherited from bases.BaseAlgorithm (private): _abc_cache, _abc_negative_cache, _abc_registry

Properties

[hide private]

Inherited from object: __class__

run(self, X, max_iter=100, start_vector=None)

source code

A kernel SVD implementation for sparse CSR matrices.

This is usually faster than np.linalg.svd when density < 20% and when
M << N or N << M (at least one order of magnitude). When M = N >= 10000
it is faster when the density < 1% and always faster regardless of
density when M = N < 10000.

These are ballpark estimates that may differ on your computer.

Parameters
----------
X : Numpy array. The matrix to decompose.

max_iter : Integer. Maximum allowed number of iterations.

start_vector : BaseStartVector. A start vector generator. Default is
        to use a random start vector.

Returns
-------
v : Numpy array. The right singular vector.

Example
-------
>>> import numpy as np
>>> from parsimony.algorithms.nipals import FastSparseSVD
>>> np.random.seed(0)
>>> X = np.random.random((10,10))
>>> fast_sparse_svd = FastSparseSVD()
>>> fast_sparse_svd.run(X)
array([[ 0.3522974 ],
       [ 0.35647707],
       [ 0.35190103],
       [ 0.34715338],
       [ 0.19594198],
       [ 0.24103104],
       [ 0.25578904],
       [ 0.29501092],
       [ 0.42311297],
       [ 0.27656382]])

Overrides: bases.ImplicitAlgorithm.run