cond

paddle.linalg. cond ( x, p=None, name=None ) [源代码]

根据范数种类 p 计算一个或一批矩阵的条件数,也可以通过 paddle.cond 来调用。

参数

  • x (Tensor):输入可以是形状为 (*, m, n) 的矩阵 Tensor, * 为零或更大的批次维度,此时 p2-2;也可以是形状为 (*, n, n) 的可逆(批)方阵 Tensor,此时 p 为任意已支持的值。数据类型为 float32 或 float64 。

  • p (float|string,可选):范数种类。目前支持的值为 fronuc1-12-2inf-inf。默认值为 None,即范数种类为 2

  • name (str,可选) - 具体用法请参见 Name,一般无需设置,默认值为 None。

返回

Tensor,条件数的计算结果,数据类型和输入 x 的一致。

代码示例

import paddle

x = paddle.to_tensor([[1., 0, -1], [0, 1, 0], [1, 0, 1]])

# compute conditional number when p is None
out = paddle.linalg.cond(x)
# Tensor(shape=[], dtype=float32, place=Place(gpu:0), stop_gradient=True,
#        1.41421342)

# compute conditional number when order of the norm is 'fro'
out_fro = paddle.linalg.cond(x, p='fro')
# Tensor(shape=[], dtype=float32, place=Place(gpu:0), stop_gradient=True,
#        3.16227770)

# compute conditional number when order of the norm is 'nuc'
out_nuc = paddle.linalg.cond(x, p='nuc')
# Tensor(shape=[], dtype=float32, place=Place(gpu:0), stop_gradient=True,
#        9.24263859)

# compute conditional number when order of the norm is 1
out_1 = paddle.linalg.cond(x, p=1)
# Tensor(shape=[], dtype=float32, place=Place(gpu:0), stop_gradient=True,
#        2.)

# compute conditional number when order of the norm is -1
out_minus_1 = paddle.linalg.cond(x, p=-1)
# Tensor(shape=[], dtype=float32, place=Place(gpu:0), stop_gradient=True,
#        1.)

# compute conditional number when order of the norm is 2
out_2 = paddle.linalg.cond(x, p=2)
# Tensor(shape=[], dtype=float32, place=Place(gpu:0), stop_gradient=True,
#        1.41421342)

# compute conditional number when order of the norm is -1
out_minus_2 = paddle.linalg.cond(x, p=-2)
# Tensor(shape=[], dtype=float32, place=Place(gpu:0), stop_gradient=True,
#        0.70710683)

# compute conditional number when order of the norm is inf
out_inf = paddle.linalg.cond(x, p=float("inf"))
# Tensor(shape=[], dtype=float32, place=Place(gpu:0), stop_gradient=True,
#        2.)

# compute conditional number when order of the norm is -inf
out_minus_inf = paddle.linalg.cond(x, p=-float("inf"))
# Tensor(shape=[], dtype=float32, place=Place(gpu:0), stop_gradient=True,
#        1.)

a = paddle.randn([2, 4, 4])
# Tensor(shape=[2, 4, 4], dtype=float32, place=Place(gpu:0), stop_gradient=True,
#        [[[-0.06784091, -0.07095790,  1.31792855, -0.58959651],
#          [ 0.20818676, -0.85640615, -0.89998871, -1.47439921],
#          [-0.49132481,  0.42250812, -0.77383220, -2.19794774],
#          [-0.33551720, -1.70003879, -1.09795380, -0.63737559]],

#         [[ 1.12026262, -0.16119350, -1.21157813,  2.74383283],
#          [-0.15999718,  0.18798758, -0.69392562,  1.35720372],
#          [-0.53013402, -2.26304483,  1.40843511, -1.02288902],
#          [ 0.69533503,  2.05261683, -0.02251151, -1.43127477]]])

a_cond_fro = paddle.linalg.cond(a, p='fro')
# Tensor(shape=[2], dtype=float32, place=Place(gpu:0), stop_gradient=True,
#        [8.86691189 , 75.23817444])

b = paddle.randn([2, 3, 4])
# Tensor(shape=[2, 3, 4], dtype=float32, place=Place(gpu:0), stop_gradient=True,
#        [[[-0.43754861,  1.80796063, -0.78729683, -1.82264030],
#          [-0.27670753,  0.06620564,  0.29072434, -0.31155765],
#          [ 0.34123746, -0.05444612,  0.05001324, -1.46877074]],

#         [[-0.64331555, -1.51103854, -1.26277697, -0.68024760],
#          [ 2.59375715, -1.06665540,  0.96575671, -0.73330832],
#          [-0.47064447, -0.23945692, -0.95150250, -1.07125998]]])
b_cond_2 = paddle.linalg.cond(b, p=2)
# Tensor(shape=[2], dtype=float32, place=Place(gpu:0), stop_gradient=True,
#        [6.64228773, 3.89068866])

使用本API的教程文档