Python numpy.linalg.qr() Examples

def test_mode_raw(self):
        # The factorization is not unique and varies between libraries,
        # so it is not possible to check against known values. Functional
        # testing is a possibility, but awaits the exposure of more
        # of the functions in lapack_lite. Consequently, this test is
        # very limited in scope. Note that the results are in FORTRAN
        # order, hence the h arrays are transposed.
        a = array([[1, 2], [3, 4], [5, 6]], dtype=np.double)

        # Test double
        h, tau = linalg.qr(a, mode='raw')
        assert_(h.dtype == np.double)
        assert_(tau.dtype == np.double)
        assert_(h.shape == (2, 3))
        assert_(tau.shape == (2,))

        h, tau = linalg.qr(a.T, mode='raw')
        assert_(h.dtype == np.double)
        assert_(tau.dtype == np.double)
        assert_(h.shape == (3, 2))
        assert_(tau.shape == (2,)) 

def _extract_factors(tens, ndims):
    """Extract iteratively the leftmost MPO tensor with given number of
    legs by a qr-decomposition

    :param np.ndarray tens: Full tensor to be factorized
    :param ndims: Number of physical legs per site or iterator over number of
        physical legs
    :returns: List of local tensors with given number of legs yielding a
        factorization of tens
    """
    current = next(ndims) if isinstance(ndims, collections.Iterator) else ndims
    if tens.ndim == current + 2:
        return [tens]
    elif tens.ndim < current + 2:
        raise AssertionError("Number of remaining legs insufficient.")
    else:
        unitary, rest = qr(tens.reshape((np.prod(tens.shape[:current + 1]), -1)))

        unitary = unitary.reshape(tens.shape[:current + 1] + rest.shape[:1])
        rest = rest.reshape(rest.shape[:1] + tens.shape[current + 1:])

        return [unitary] + _extract_factors(rest, ndims) 

def _lcanonicalize(self, to_site):
        """Right-canonicalizes all local tensors _ltens[to_site:] in place

        :param to_site: Index of the site up to which canonicalization is to be
            performed

        """
        assert 0 < to_site <= len(self), 'to_site={!r}'.format(to_site)

        lcanon, rcanon = self.canonical_form
        for site in range(rcanon - 1, to_site - 1, -1):
            ltens = self._lt[site]
            q, r = qr(ltens.reshape((ltens.shape[0], -1)).T)
            # if ltens.shape[-1] > prod(ltens.phys_shape) --> trivial comp.
            # can be accounted by adapting rank here
            newtens = (matdot(self._lt[site - 1], r.T),
                       q.T.reshape((-1,) + ltens.shape[1:]))
            self._lt.update(slice(site - 1, site + 1), newtens,
                            canonicalization=(None, 'right')) 

def _rcanonicalize(self, to_site):
        """Left-canonicalizes all local tensors _ltens[:to_site] in place

        :param to_site: Index of the site up to which canonicalization is to be
            performed

        """
        assert 0 <= to_site < len(self), 'to_site={!r}'.format(to_site)

        lcanon, rcanon = self._lt.canonical_form
        for site in range(lcanon, to_site):
            ltens = self._lt[site]
            q, r = qr(ltens.reshape((-1, ltens.shape[-1])))
            # if ltens.shape[-1] > prod(ltens.phys_shape) --> trivial comp.
            # can be accounted by adapting rank here
            newtens = (q.reshape(ltens.shape[:-1] + (-1,)),
                       matdot(r, self._lt[site + 1]))
            self._lt.update(slice(site, site + 2), newtens,
                            canonicalization=('left', None)) 

def test_mode_raw(self):
        # The factorization is not unique and varies between libraries,
        # so it is not possible to check against known values. Functional
        # testing is a possibility, but awaits the exposure of more
        # of the functions in lapack_lite. Consequently, this test is
        # very limited in scope. Note that the results are in FORTRAN
        # order, hence the h arrays are transposed.
        a = array([[1, 2], [3, 4], [5, 6]], dtype=np.double)

        # Test double
        h, tau = linalg.qr(a, mode='raw')
        assert_(h.dtype == np.double)
        assert_(tau.dtype == np.double)
        assert_(h.shape == (2, 3))
        assert_(tau.shape == (2,))

        h, tau = linalg.qr(a.T, mode='raw')
        assert_(h.dtype == np.double)
        assert_(tau.dtype == np.double)
        assert_(h.shape == (3, 2))
        assert_(tau.shape == (2,)) 

def test_mode_raw(self):
        # The factorization is not unique and varies between libraries,
        # so it is not possible to check against known values. Functional
        # testing is a possibility, but awaits the exposure of more
        # of the functions in lapack_lite. Consequently, this test is
        # very limited in scope. Note that the results are in FORTRAN
        # order, hence the h arrays are transposed.
        a = self.array([[1, 2], [3, 4], [5, 6]], dtype=np.double)

        # Test double
        h, tau = linalg.qr(a, mode='raw')
        assert_(h.dtype == np.double)
        assert_(tau.dtype == np.double)
        assert_(h.shape == (2, 3))
        assert_(tau.shape == (2,))

        h, tau = linalg.qr(a.T, mode='raw')
        assert_(h.dtype == np.double)
        assert_(tau.dtype == np.double)
        assert_(h.shape == (3, 2))
        assert_(tau.shape == (2,)) 

def test_mode_raw(self):
        # The factorization is not unique and varies between libraries,
        # so it is not possible to check against known values. Functional
        # testing is a possibility, but awaits the exposure of more
        # of the functions in lapack_lite. Consequently, this test is
        # very limited in scope. Note that the results are in FORTRAN
        # order, hence the h arrays are transposed.
        a = self.array([[1, 2], [3, 4], [5, 6]], dtype=np.double)

        # Test double
        h, tau = linalg.qr(a, mode='raw')
        assert_(h.dtype == np.double)
        assert_(tau.dtype == np.double)
        assert_(h.shape == (2, 3))
        assert_(tau.shape == (2,))

        h, tau = linalg.qr(a.T, mode='raw')
        assert_(h.dtype == np.double)
        assert_(tau.dtype == np.double)
        assert_(h.shape == (3, 2))
        assert_(tau.shape == (2,)) 

def test_mode_raw(self):
        # The factorization is not unique and varies between libraries,
        # so it is not possible to check against known values. Functional
        # testing is a possibility, but awaits the exposure of more
        # of the functions in lapack_lite. Consequently, this test is
        # very limited in scope. Note that the results are in FORTRAN
        # order, hence the h arrays are transposed.
        a = self.array([[1, 2], [3, 4], [5, 6]], dtype=np.double)

        # Test double
        h, tau = linalg.qr(a, mode='raw')
        assert_(h.dtype == np.double)
        assert_(tau.dtype == np.double)
        assert_(h.shape == (2, 3))
        assert_(tau.shape == (2,))

        h, tau = linalg.qr(a.T, mode='raw')
        assert_(h.dtype == np.double)
        assert_(tau.dtype == np.double)
        assert_(h.shape == (3, 2))
        assert_(tau.shape == (2,)) 

def test_mode_raw(self):
        # The factorization is not unique and varies between libraries,
        # so it is not possible to check against known values. Functional
        # testing is a possibility, but awaits the exposure of more
        # of the functions in lapack_lite. Consequently, this test is
        # very limited in scope. Note that the results are in FORTRAN
        # order, hence the h arrays are transposed.
        a = self.array([[1, 2], [3, 4], [5, 6]], dtype=np.double)

        # Test double
        h, tau = linalg.qr(a, mode='raw')
        assert_(h.dtype == np.double)
        assert_(tau.dtype == np.double)
        assert_(h.shape == (2, 3))
        assert_(tau.shape == (2,))

        h, tau = linalg.qr(a.T, mode='raw')
        assert_(h.dtype == np.double)
        assert_(tau.dtype == np.double)
        assert_(h.shape == (3, 2))
        assert_(tau.shape == (2,)) 

def test_mode_raw(self):
        # The factorization is not unique and varies between libraries,
        # so it is not possible to check against known values. Functional
        # testing is a possibility, but awaits the exposure of more
        # of the functions in lapack_lite. Consequently, this test is
        # very limited in scope. Note that the results are in FORTRAN
        # order, hence the h arrays are transposed.
        a = array([[1, 2], [3, 4], [5, 6]], dtype=np.double)
        b = a.astype(np.single)

        # Test double
        h, tau = linalg.qr(a, mode='raw')
        assert_(h.dtype == np.double)
        assert_(tau.dtype == np.double)
        assert_(h.shape == (2, 3))
        assert_(tau.shape == (2,))

        h, tau = linalg.qr(a.T, mode='raw')
        assert_(h.dtype == np.double)
        assert_(tau.dtype == np.double)
        assert_(h.shape == (3, 2))
        assert_(tau.shape == (2,)) 

def test_mode_raw(self):
        # The factorization is not unique and varies between libraries,
        # so it is not possible to check against known values. Functional
        # testing is a possibility, but awaits the exposure of more
        # of the functions in lapack_lite. Consequently, this test is
        # very limited in scope. Note that the results are in FORTRAN
        # order, hence the h arrays are transposed.
        a = self.array([[1, 2], [3, 4], [5, 6]], dtype=np.double)

        # Test double
        h, tau = linalg.qr(a, mode='raw')
        assert_(h.dtype == np.double)
        assert_(tau.dtype == np.double)
        assert_(h.shape == (2, 3))
        assert_(tau.shape == (2,))

        h, tau = linalg.qr(a.T, mode='raw')
        assert_(h.dtype == np.double)
        assert_(tau.dtype == np.double)
        assert_(h.shape == (3, 2))
        assert_(tau.shape == (2,)) 

def test_mode_raw(self):
        # The factorization is not unique and varies between libraries,
        # so it is not possible to check against known values. Functional
        # testing is a possibility, but awaits the exposure of more
        # of the functions in lapack_lite. Consequently, this test is
        # very limited in scope. Note that the results are in FORTRAN
        # order, hence the h arrays are transposed.
        a = self.array([[1, 2], [3, 4], [5, 6]], dtype=np.double)

        # Test double
        h, tau = linalg.qr(a, mode='raw')
        assert_(h.dtype == np.double)
        assert_(tau.dtype == np.double)
        assert_(h.shape == (2, 3))
        assert_(tau.shape == (2,))

        h, tau = linalg.qr(a.T, mode='raw')
        assert_(h.dtype == np.double)
        assert_(tau.dtype == np.double)
        assert_(h.shape == (3, 2))
        assert_(tau.shape == (2,)) 

def test_mode_raw(self):
        # The factorization is not unique and varies between libraries,
        # so it is not possible to check against known values. Functional
        # testing is a possibility, but awaits the exposure of more
        # of the functions in lapack_lite. Consequently, this test is
        # very limited in scope. Note that the results are in FORTRAN
        # order, hence the h arrays are transposed.
        a = array([[1, 2], [3, 4], [5, 6]], dtype=np.double)

        # Test double
        h, tau = linalg.qr(a, mode='raw')
        assert_(h.dtype == np.double)
        assert_(tau.dtype == np.double)
        assert_(h.shape == (2, 3))
        assert_(tau.shape == (2,))

        h, tau = linalg.qr(a.T, mode='raw')
        assert_(h.dtype == np.double)
        assert_(tau.dtype == np.double)
        assert_(h.shape == (3, 2))
        assert_(tau.shape == (2,)) 

def test_mode_raw(self):
        # The factorization is not unique and varies between libraries,
        # so it is not possible to check against known values. Functional
        # testing is a possibility, but awaits the exposure of more
        # of the functions in lapack_lite. Consequently, this test is
        # very limited in scope. Note that the results are in FORTRAN
        # order, hence the h arrays are transposed.
        a = self.array([[1, 2], [3, 4], [5, 6]], dtype=np.double)

        # Test double
        h, tau = linalg.qr(a, mode='raw')
        assert_(h.dtype == np.double)
        assert_(tau.dtype == np.double)
        assert_(h.shape == (2, 3))
        assert_(tau.shape == (2,))

        h, tau = linalg.qr(a.T, mode='raw')
        assert_(h.dtype == np.double)
        assert_(tau.dtype == np.double)
        assert_(h.shape == (3, 2))
        assert_(tau.shape == (2,)) 

def test_mode_raw(self):
        # The factorization is not unique and varies between libraries,
        # so it is not possible to check against known values. Functional
        # testing is a possibility, but awaits the exposure of more
        # of the functions in lapack_lite. Consequently, this test is
        # very limited in scope. Note that the results are in FORTRAN
        # order, hence the h arrays are transposed.
        a = array([[1, 2], [3, 4], [5, 6]], dtype=np.double)

        # Test double
        h, tau = linalg.qr(a, mode='raw')
        assert_(h.dtype == np.double)
        assert_(tau.dtype == np.double)
        assert_(h.shape == (2, 3))
        assert_(tau.shape == (2,))

        h, tau = linalg.qr(a.T, mode='raw')
        assert_(h.dtype == np.double)
        assert_(tau.dtype == np.double)
        assert_(h.shape == (3, 2))
        assert_(tau.shape == (2,)) 

def test_mode_raw(self):
        # The factorization is not unique and varies between libraries,
        # so it is not possible to check against known values. Functional
        # testing is a possibility, but awaits the exposure of more
        # of the functions in lapack_lite. Consequently, this test is
        # very limited in scope. Note that the results are in FORTRAN
        # order, hence the h arrays are transposed.
        a = array([[1, 2], [3, 4], [5, 6]], dtype=np.double)
        b = a.astype(np.single)

        # Test double
        h, tau = linalg.qr(a, mode='raw')
        assert_(h.dtype == np.double)
        assert_(tau.dtype == np.double)
        assert_(h.shape == (2, 3))
        assert_(tau.shape == (2,))

        h, tau = linalg.qr(a.T, mode='raw')
        assert_(h.dtype == np.double)
        assert_(tau.dtype == np.double)
        assert_(h.shape == (3, 2))
        assert_(tau.shape == (2,)) 

def test_mode_raw(self):
        # The factorization is not unique and varies between libraries,
        # so it is not possible to check against known values. Functional
        # testing is a possibility, but awaits the exposure of more
        # of the functions in lapack_lite. Consequently, this test is
        # very limited in scope. Note that the results are in FORTRAN
        # order, hence the h arrays are transposed.
        a = self.array([[1, 2], [3, 4], [5, 6]], dtype=np.double)

        # Test double
        h, tau = linalg.qr(a, mode='raw')
        assert_(h.dtype == np.double)
        assert_(tau.dtype == np.double)
        assert_(h.shape == (2, 3))
        assert_(tau.shape == (2,))

        h, tau = linalg.qr(a.T, mode='raw')
        assert_(h.dtype == np.double)
        assert_(tau.dtype == np.double)
        assert_(h.shape == (3, 2))
        assert_(tau.shape == (2,)) 

def test_mode_raw(self):
        # The factorization is not unique and varies between libraries,
        # so it is not possible to check against known values. Functional
        # testing is a possibility, but awaits the exposure of more
        # of the functions in lapack_lite. Consequently, this test is
        # very limited in scope. Note that the results are in FORTRAN
        # order, hence the h arrays are transposed.
        a = array([[1, 2], [3, 4], [5, 6]], dtype=np.double)

        # Test double
        h, tau = linalg.qr(a, mode='raw')
        assert_(h.dtype == np.double)
        assert_(tau.dtype == np.double)
        assert_(h.shape == (2, 3))
        assert_(tau.shape == (2,))

        h, tau = linalg.qr(a.T, mode='raw')
        assert_(h.dtype == np.double)
        assert_(tau.dtype == np.double)
        assert_(h.shape == (3, 2))
        assert_(tau.shape == (2,)) 

def test_mode_raw(self):
        # The factorization is not unique and varies between libraries,
        # so it is not possible to check against known values. Functional
        # testing is a possibility, but awaits the exposure of more
        # of the functions in lapack_lite. Consequently, this test is
        # very limited in scope. Note that the results are in FORTRAN
        # order, hence the h arrays are transposed.
        a = array([[1, 2], [3, 4], [5, 6]], dtype=np.double)

        # Test double
        h, tau = linalg.qr(a, mode='raw')
        assert_(h.dtype == np.double)
        assert_(tau.dtype == np.double)
        assert_(h.shape == (2, 3))
        assert_(tau.shape == (2,))

        h, tau = linalg.qr(a.T, mode='raw')
        assert_(h.dtype == np.double)
        assert_(tau.dtype == np.double)
        assert_(h.shape == (3, 2))
        assert_(tau.shape == (2,)) 

def __init__(self):
        matrix_pairs = {}
        rng = np.random.RandomState(SEED)
        for sz in SIZES:
            eigvec = nla.qr(rng.standard_normal((sz, sz)))[0]
            eigval = np.abs(rng.standard_normal(sz))
            matrix_pairs[sz] = (
                matrices.EigendecomposedPositiveDefiniteMatrix(eigvec, eigval),
                (eigvec * eigval) @ eigvec.T)
        super().__init__(matrix_pairs, rng) 

def test_nystrom_extension(seed=123):
    """ Test Nystrom Extension: low rank approximation is exact when
    G is itself low rank
    """
    n = 10
    s = 2
    rng = np.random.RandomState(seed)
    X = rng.randn(n, s)
    G = np.dot(X, X.T) # has rank s

    # find the linearly independent columns of 
    q = qr(G)[1] 
    q = absolute(q)
    sums = np.sum(q,axis=1)
    i = 0
    dims = list()
    while( i < n ): #dim is the matrix dimension
        if(sums[i] > 1.e-10):
            dims.append(i)
        i += 1
    
    # Find the eigendecomposition of the full rank portion:
    W = G[dims,:]
    W = W[:,dims]
    eval, evec = np.linalg.eigh(W)
    
    # pass the dims columns of G 
    C = G[:,dims]
    # Find the estimated eigendecomposition using Nystrom 
    eval_nystrom, evec_nystrom = nystrom_extension(C, evec, eval)
        
    # reconstruct G using Nystrom Approximatiuon 
    G_nystrom = np.dot(np.dot(evec_nystrom, np.diag(eval_nystrom)),evec_nystrom.T)
    # since rank(W) = rank(G) = s the nystrom approximation of G is exact:
    assert_array_almost_equal(G_nystrom, G) 

def __init__(self):
        matrix_pairs = {}
        rng = np.random.RandomState(SEED)
        for sz in SIZES:
            array = nla.qr(rng.standard_normal((sz, sz)))[0]
            matrix_pairs[sz] = (matrices.OrthogonalMatrix(array), array)
            super().__init__(matrix_pairs, rng) 

def __init__(self):
        matrix_pairs = {}
        rng = np.random.RandomState(SEED)
        for sz in SIZES:
            eigvec = nla.qr(rng.standard_normal((sz, sz)))[0]
            eigval = rng.standard_normal(sz)
            matrix_pairs[sz] = (
                matrices.EigendecomposedSymmetricMatrix(eigvec, eigval),
                (eigvec * eigval) @ eigvec.T)
        super().__init__(matrix_pairs, rng) 

def test_qr_empty(self):
        a = np.zeros((0, 2))
        assert_raises(linalg.LinAlgError, linalg.qr, a) 

def test_0_size(self):
        # There may be good ways to do (some of this) reasonably:
        a = np.zeros((0, 0))
        assert_raises(linalg.LinAlgError, linalg.qr, a)
        a = np.zeros((0, 1))
        assert_raises(linalg.LinAlgError, linalg.qr, a)
        a = np.zeros((1, 0))
        assert_raises(linalg.LinAlgError, linalg.qr, a) 

def __init__(self):
        matrix_pairs = {}
        rng = np.random.RandomState(SEED)
        for sz in SIZES:
            orth_array = nla.qr(rng.standard_normal((sz, sz)))[0]
            scalar = rng.standard_normal()
            matrix_pairs[sz] = (
                matrices.ScaledOrthogonalMatrix(scalar, orth_array),
                scalar * orth_array)
            super().__init__(matrix_pairs, rng) 

def check_qr(self, a):
        # This test expects the argument `a` to be an ndarray or
        # a subclass of an ndarray of inexact type.
        a_type = type(a)
        a_dtype = a.dtype
        m, n = a.shape
        k = min(m, n)

        # mode == 'complete'
        q, r = linalg.qr(a, mode='complete')
        assert_(q.dtype == a_dtype)
        assert_(r.dtype == a_dtype)
        assert_(isinstance(q, a_type))
        assert_(isinstance(r, a_type))
        assert_(q.shape == (m, m))
        assert_(r.shape == (m, n))
        assert_almost_equal(dot(q, r), a)
        assert_almost_equal(dot(q.T.conj(), q), np.eye(m))
        assert_almost_equal(np.triu(r), r)

        # mode == 'reduced'
        q1, r1 = linalg.qr(a, mode='reduced')
        assert_(q1.dtype == a_dtype)
        assert_(r1.dtype == a_dtype)
        assert_(isinstance(q1, a_type))
        assert_(isinstance(r1, a_type))
        assert_(q1.shape == (m, k))
        assert_(r1.shape == (k, n))
        assert_almost_equal(dot(q1, r1), a)
        assert_almost_equal(dot(q1.T.conj(), q1), np.eye(k))
        assert_almost_equal(np.triu(r1), r1)

        # mode == 'r'
        r2 = linalg.qr(a, mode='r')
        assert_(r2.dtype == a_dtype)
        assert_(isinstance(r2, a_type))
        assert_almost_equal(r2, r1) 

def check_qr(self, a):
        # This test expects the argument `a` to be an ndarray or
        # a subclass of an ndarray of inexact type.
        a_type = type(a)
        a_dtype = a.dtype
        m, n = a.shape
        k = min(m, n)

        # mode == 'complete'
        q, r = linalg.qr(a, mode='complete')
        assert_(q.dtype == a_dtype)
        assert_(r.dtype == a_dtype)
        assert_(isinstance(q, a_type))
        assert_(isinstance(r, a_type))
        assert_(q.shape == (m, m))
        assert_(r.shape == (m, n))
        assert_almost_equal(dot(q, r), a)
        assert_almost_equal(dot(q.T.conj(), q), np.eye(m))
        assert_almost_equal(np.triu(r), r)

        # mode == 'reduced'
        q1, r1 = linalg.qr(a, mode='reduced')
        assert_(q1.dtype == a_dtype)
        assert_(r1.dtype == a_dtype)
        assert_(isinstance(q1, a_type))
        assert_(isinstance(r1, a_type))
        assert_(q1.shape == (m, k))
        assert_(r1.shape == (k, n))
        assert_almost_equal(dot(q1, r1), a)
        assert_almost_equal(dot(q1.T.conj(), q1), np.eye(k))
        assert_almost_equal(np.triu(r1), r1)

        # mode == 'r'
        r2 = linalg.qr(a, mode='r')
        assert_(r2.dtype == a_dtype)
        assert_(isinstance(r2, a_type))
        assert_almost_equal(r2, r1) 

def test_0_size(self):
        # There may be good ways to do (some of this) reasonably:
        a = np.zeros((0, 0))
        assert_raises(linalg.LinAlgError, linalg.qr, a)
        a = np.zeros((0, 1))
        assert_raises(linalg.LinAlgError, linalg.qr, a)
        a = np.zeros((1, 0))
        assert_raises(linalg.LinAlgError, linalg.qr, a) 

def check_qr(self, a):
        # This test expects the argument `a` to be an ndarray or
        # a subclass of an ndarray of inexact type.
        a_type = type(a)
        a_dtype = a.dtype
        m, n = a.shape
        k = min(m, n)

        # mode == 'complete'
        q, r = linalg.qr(a, mode='complete')
        assert_(q.dtype == a_dtype)
        assert_(r.dtype == a_dtype)
        assert_(isinstance(q, a_type))
        assert_(isinstance(r, a_type))
        assert_(q.shape == (m, m))
        assert_(r.shape == (m, n))
        assert_almost_equal(dot(q, r), a)
        assert_almost_equal(dot(q.T.conj(), q), np.eye(m))
        assert_almost_equal(np.triu(r), r)

        # mode == 'reduced'
        q1, r1 = linalg.qr(a, mode='reduced')
        assert_(q1.dtype == a_dtype)
        assert_(r1.dtype == a_dtype)
        assert_(isinstance(q1, a_type))
        assert_(isinstance(r1, a_type))
        assert_(q1.shape == (m, k))
        assert_(r1.shape == (k, n))
        assert_almost_equal(dot(q1, r1), a)
        assert_almost_equal(dot(q1.T.conj(), q1), np.eye(k))
        assert_almost_equal(np.triu(r1), r1)

        # mode == 'r'
        r2 = linalg.qr(a, mode='r')
        assert_(r2.dtype == a_dtype)
        assert_(isinstance(r2, a_type))
        assert_almost_equal(r2, r1)