[英]Numpy error: shape mismatch
When I was trying to solve a scientific problem with Python (Numpy), a 'shape mismatch' error came up: "shape mismatch: objects cannot be broadcast to a single shape". 当我试图用Python(Numpy)解决科学问题时,出现了“形状不匹配”错误:“形状不匹配:对象无法广播到单个形状”。 I managed to reproduce the same error in a simpler form, as shown below: 我设法以更简单的形式重现相同的错误,如下所示:
import numpy as np
nx = 3; ny = 5
ff = np.ones([nx,ny,7])
def test(x, y):
z = 0.0
for i in range(7):
z = z + ff[x,y,i]
return z
print test(np.arange(nx),np.arange(ny))
When I tried to call test(x,y)
with x=1,y=np.arange(ny)
, everything works fine. 当我试图用x=1,y=np.arange(ny)
调用test(x,y)
时,一切正常。 So what's going on here? 那么这里发生了什么? Why can't the both parameters be numpy arrays? 为什么两个参数都不能成为numpy数组?
UPDATE UPDATE
I have worked out the problem with some hints from @Saullo Castro. 我用@Saullo Castro的一些提示解决了这个问题。 Here's some updated info for you guys who tried to help but feel unclear about my intention: 这里有一些更新的信息给那些试图帮助但不清楚我的意图的人:
Basically I created a mesh grid with dimension nx*ny and another array ff
that stores some value for each node. 基本上我创建了一个维度为nx * ny的网格网格和另一个为每个节点存储一些值的数组ff
。 In the above code, ff
has 7 values for each node and I was trying to sum up the 7 values to get a new nx*ny array. 在上面的代码中, ff
为每个节点有7个值,我试图总结7个值以获得一个新的nx * ny数组。
However, the "shape mismatch" error is not due to the summing process as many of you might have guess now. 但是,“形状不匹配”错误并不是由于求和过程,因为许多人现在可能已经猜到了。 I have misunderstood the rule of functions taking ndarray objects as input parameters. 我误解了将ndarray对象作为输入参数的函数规则。 I tried to pass np.arange(nx), np.arange(ny)
to test()
is not gonna give me what I desired, even if nx==ny
. 我试图传递np.arange(nx), np.arange(ny)
到test()
不会给我我想要的东西,即使nx==ny
。
Back to my original intention, I solve the problem by creating another function and used np.fromfunction
to created the array: 回到我的初衷,我通过创建另一个函数并使用np.fromfunction
来创建数组来解决问题:
def tt(x, y):
return np.fromfunction(lambda a,b: test(a,b), (x, y))
which is not perfect but it works. 这不完美,但它的工作原理。 (In this example there seems to be no need to create a new function, but in my actual code I modified it a bit so it can be used for slice of the grid) (在这个例子中,似乎没有必要创建一个新函数,但在我的实际代码中我修改了一下,因此它可以用于网格的切片)
Anyway, I do believe there's a much better way compared to my kind of dirty solution. 无论如何,我相信与我的那种肮脏的解决方案相比,有更好的方法。 So if you have any idea about that, please share with us :). 所以,如果您对此有任何想法,请与我们分享:)。
Let's look into an array similar to your ff
array: 让我们看一下类似于你的ff
数组的数组:
nx = 3; ny = 4
ff = np.arange(nx*ny*5).reshape(nx,ny,5)
#array([[[ 0, 1, 2, 3, 4],
# [ 5, 6, 7, 8, 9],
# [10, 11, 12, 13, 14],
# [15, 16, 17, 18, 19]],
#
# [[20, 21, 22, 23, 24],
# [25, 26, 27, 28, 29],
# [30, 31, 32, 33, 34],
# [35, 36, 37, 38, 39]],
#
# [[40, 41, 42, 43, 44],
# [45, 46, 47, 48, 49],
# [50, 51, 52, 53, 54],
# [55, 56, 57, 58, 59]]])
When you index using arrays of indices a, b, c
like in ff[a, b, c]
, a, b, c
must have the same shape, and numpy
will build a new array based on the indices. 使用索引a, b, c
索引进行索引时a, b, c
如ff[a, b, c]
, a, b, c
必须具有相同的形状, numpy
将根据索引构建一个新数组。 For example: 例如:
ff[[0, 0, 1, 1, 2, 2], [0, 1, 0, 1, 2, 3], [0, 0, 0, 1, 1, 1]]
#array([ 0, 5, 20, 26, 51, 56])
This is called fancy indexing, which is like building an array with: 这称为花式索引,就像构建一个数组:
np.array([ff[0, 0, 0], ff[0, 1, 0], ff[1, 0, 0], ..., ff[2, 3, 1]])
In your case the f[x, y, i]
will produce a shape mismatch error since a, b, c
do not have the same shape. 在您的情况下, f[x, y, i]
将产生形状不匹配错误,因为a, b, c
不具有相同的形状。
Looks like you want to sum ff
over the last dimension, with the 1st 2 dimensions covering their whole range. 看起来你想要在最后一个维度上加总ff
,其中前两个维度涵盖了它们的整个范围。 :
is used to denote the whole range of a dimension: :
用于表示维度的整个范围:
def test():
z = 0.0
for i in range(7):
z = z + ff[:,:,i]
return z
print test()
But you can get the same result without looping, by using the sum
method. 但是,通过使用sum
方法,您可以在不循环的情况下获得相同的结果。
print ff.sum(axis=-1)
:
is shorthand for 0:n
:
是0:n
简写
ff[0:nx, 0:ny, 0]==ff[:,:,0]
It is possible to index a block of ff
with ranges, but you have to be much more careful about the shapes of the indexing arrays. 可以使用范围索引ff
块,但是您必须更加小心索引数组的形状。 For a beginner it is better to focus on getting slicing
and broadcasting
correct. 对于初学者来说,最好将注意力集中在slicing
和broadcasting
上。
edit - 编辑 -
You can index an array like ff
with arrays generated by meshgrid
: 您可以使用meshgrid
生成的数组索引像ff
的数组:
I,J = meshgrid(np.arange(nx),np.arange(ny),indexing='ij',sparse=False)
I.shape # (nx,ny)
ff[I,J,:]
also works with 也适用
I,J = meshgrid(np.arange(nx),np.arange(ny),indexing='ij',sparse=True)
I.shape # (nx,1)
J.shape # (1, ny)
ogrid
and mgrid
are alternatives to meshgrid
. ogrid
和mgrid
是meshgrid
替代meshgrid
。
Let's reproduce your problem in 2D case, so it is easier to see: 让我们在2D情况下重现您的问题,因此更容易看到:
import numpy as np
a = np.arange(15).reshape(3,5)
x = np.arange(3)
y = np.arange(5)
Demo: 演示:
>>> a
array([[ 0, 1, 2, 3, 4],
[ 5, 6, 7, 8, 9],
[10, 11, 12, 13, 14]])
>>> a[x, y] # <- This is the error that you are getting
Traceback (most recent call last):
File "<stdin>", line 1, in <module>
ValueError: shape mismatch: objects cannot be broadcast to a single shape
# You are getting the error because x and y are different lengths,
# If x and y were the same lengths, the code would work:
>>> a[x, x]
array([ 0, 6, 12])
# mixing arrays and scalars is not a problem
>>> a[x, 2]
array([ 2, 7, 12])
It is not clear in your question what you are trying to do or what result are you expecting. 在你的问题中,你不清楚你想要做什么或你期望什么结果。 It seems, though, that you are trying to calculate a total with your variable z
. 但是,似乎您正在尝试使用变量z
计算总计。
Check if the sum
method produces the result that you need: 检查sum
方法是否产生您需要的结果:
import numpy as np
nx = 3; ny = 5
ff = ff = np.array(np.arange(nx*ny*7)).reshape(nx,ny,7)
print ff.sum() # 5460
print ff.sum(axis=0) # array([[105, 108, 111, 114, 117, 120, 123],
# [126, 129, 132, 135, 138, 141, 144],
# [147, 150, 153, 156, 159, 162, 165],
# [168, 171, 174, 177, 180, 183, 186],
# [189, 192, 195, 198, 201, 204, 207]]) shape(5,7)
print ff.sum(axis=1) # array([[ 70, 75, 80, 85, 90, 95, 100],
# [245, 250, 255, 260, 265, 270, 275],
# [420, 425, 430, 435, 440, 445, 450]]) shape (3,7)
print ff.sum(axis=2) # array([[ 21, 70, 119, 168, 217],
# [266, 315, 364, 413, 462],
# [511, 560, 609, 658, 707]]) shape (3,5)
声明:本站的技术帖子网页,遵循CC BY-SA 4.0协议,如果您需要转载,请注明本站网址或者原文地址。任何问题请咨询:yoyou2525@163.com.