Esempio n. 1
0
def _randomized_svd(M, k, p, q, random_state):
    random_state = check_random_state(random_state)

    Q = randomized_range_finder(M, k + p, q, random_state)

    # project M to the (k + p) dimensional space using the basis vectors
    B = safe_sparse_dot(Q.T, M)

    # compute the SVD on the thin matrix: (k + p) wide
    Uhat_t, S, Vt = svd_ndarray(B, k)
    del B
    U = np.dot(Q, Uhat_t.T)

    return U, S, Vt.T
Esempio n. 2
0
def _randomized_svd(M, k, p, q, random_state):
    random_state = check_random_state(random_state)
    
    Q = randomized_range_finder(M, k+p, q, random_state)

    # project M to the (k + p) dimensional space using the basis vectors
    B = safe_sparse_dot(Q.T, M)

    # compute the SVD on the thin matrix: (k + p) wide
    Uhat_t, S, Vt = svd_ndarray(B, k)
    del B
    U = np.dot(Q, Uhat_t.T)

    return U, S, Vt.T
Esempio n. 3
0
def svd(matrix, k=50):
    """
    Calculate the truncated singular value decomposition
    :math:`A = U * Sigma * V^T` using SVDLIBC.

    Returns a triple of:
    
    - U as a dense labeled matrix
    - S, a dense vector representing the diagonal of Sigma
    - V as a dense labeled matrix
    
    This matrix must not contain any empty rows or columns. If it does,
    use the .squish() method first.
    """
    assert matrix.ndim == 2
    if isinstance(matrix, DenseMatrix):
        Ut, S, Vt = svd_ndarray(matrix, k)
    elif isinstance(matrix, SparseMatrix):
        if matrix.nnz == 0:
            # don't let svdlib touch a matrix of all zeros. It explodes and
            # corrupts its state. Just return a zero result instead.
            U = DenseMatrix((matrix.shape[0], k))
            S = np.zeros((k,))
            V = DenseMatrix((matrix.shape[1], k))
            return U, S, V
        if matrix.shape[1] >= matrix.shape[0] * 1.2:
            # transpose the matrix for speed
            V, S, U = matrix.T.svd(k)
            return U, S, V
        Ut, S, Vt = svd_llmat(matrix.llmatrix, k)
    else:
        raise TypeError("Don't know how to SVD a %r", type(matrix))

    U = DenseMatrix(Ut.T, matrix.row_labels, None)
    V = DenseMatrix(Vt.T, matrix.col_labels, None)

    return U, S, V
Esempio n. 4
0
def svd(matrix, k=50):
    """
    Calculate the truncated singular value decomposition
    :math:`A = U * Sigma * V^T` using SVDLIBC.

    Returns a triple of:
    
    - U as a dense labeled matrix
    - S, a dense vector representing the diagonal of Sigma
    - V as a dense labeled matrix
    
    This matrix must not contain any empty rows or columns. If it does,
    use the .squish() method first.
    """
    assert matrix.ndim == 2
    if isinstance(matrix, DenseMatrix):
        Ut, S, Vt = svd_ndarray(matrix, k)
    elif isinstance(matrix, SparseMatrix):
        if matrix.nnz == 0:
            # don't let svdlib touch a matrix of all zeros. It explodes and
            # corrupts its state. Just return a zero result instead.
            U = DenseMatrix((matrix.shape[0], k))
            S = np.zeros((k, ))
            V = DenseMatrix((matrix.shape[1], k))
            return U, S, V
        if matrix.shape[1] >= matrix.shape[0] * 1.2:
            # transpose the matrix for speed
            V, S, U = matrix.T.svd(k)
            return U, S, V
        Ut, S, Vt = svd_llmat(matrix.llmatrix, k)
    else:
        raise TypeError("Don't know how to SVD a %r", type(matrix))

    U = DenseMatrix(Ut.T, matrix.row_labels, None)
    V = DenseMatrix(Vt.T, matrix.col_labels, None)

    return U, S, V