NumPy 对可变大小、多维数组进行切片

Question

假设有以下几行代码

import numpy as np

# The values equal to 1 inside this nested list indicate where the data need to be loaded. a = [7 x 6]
a = [
    [0, 1, 0, 1, None, None],
    [0, 0, 0, 0, None, 0],
    [0, 0, 1, 0, None, 0],
    [0, 1, 0, 1, None, 1],
    [0, 0, 0, 1, None, 0],
    [0, 0, 0, 0, None, 0],
    [1, 0, 0, 0, None, None]
]
# The list "a" cannot be modified for a number of reasons, so I create a np.array copy, named "b"
b = np.array(a)

N = int(1E7)  # Number of samples

# The loop below retrieves the positions inside "b" in which data need to be loaded
row = []
col = []
for i in range(len(b)):
    col.append([])
    if any(b[i] == 1):
        row.append(i)
    for j in range(len(a[i])):
        if b[i][j] is 1:
            b[i][j] = np.zeros((N, 1))
            col[i].append(j)


# Loading the data inside the selected positions of "b". "mydata" is a numpy array, whose shape is (N, 6)
for i in row:
    mydata = np.random.randn(N, len(a[0])).reshape(N, len(a[0])) # Generation of dummy data
    b[i, col[i]] = mydata[:, col[i]]  # This instruction returns a ValueError

但是，我收到以下错误： ValueError：形状不匹配：形状（10000000,2）的值数组无法广播到形状（2，）的索引结果

为什么这种切片不能正常工作？是不是因为"b"里面的数组元素大小可变？

提前谢谢你。

Answer 1

对于由于使用b[i, col[i]]而在行中插入多个零数组的情况，切片无法正常工作。

只需考虑您的第一行。这给了你row=[0]，col =[[1,3]]。这意味着b[0,0] 引用了第 1 列 和 3 的 zeros 数组。您应该像以前一样使用嵌套的 for 循环遍历行和列

for i in row:
    for j in col[i]:
        mydata = np.random.randn(N, len(a[0])).reshape(N, len(a[0]))
        b[i, j] = mydata[:, col[i]]

Answer 2

让我们将N 简化为合理的值，并添加一些打印：

print(row)
print(col)
print(a)
print(b)

跑步：

0942:~/mypy$ python3 stack60813103.py 
[0, 2, 3, 4, 6]
[[1, 3], [], [2], [1, 3, 5], [3], [], [0]]
[[0, 1, 0, 1, None, None], [0, 0, 0, 0, None, 0], [0, 0, 1, 0, None, 0], [0, 1, 0, 1, None, 1], [0, 0, 0, 1, None, 0], [0, 0, 0, 0, None, 0], [1, 0, 0, 0, None, None]]
[[0 array([[0.],
       [0.],
       [0.],
       [0.],
       [0.]]) 0
  array([[0.],
       [0.],
       [0.],
       [0.],
       [0.]]) None
  None]
 [0 0 0 0 None 0]
 [0 0 array([[0.],
       [0.],
       [0.],
       [0.],
       [0.]]) 0
  None 0]
 ....
 [0 0 0 0 None 0]
 [array([[0.],
       [0.],
       [0.],
       [0.],
       [0.]]) 0 0 0
  None None]]
Traceback (most recent call last):
  File "stack60813103.py", line 38, in <module>
    b[i, col[i]] = mydata[:, col[i]]  # This instruction returns a ValueError
ValueError: shape mismatch: value array of shape (5,2) could not be broadcast to indexing result of shape (2,)

row、col 和 a 是列表，b 是对象 dtype 数组（因为所有 None）。您的循环插入了一堆 np.zeros((N,1)) 数组。

mydata 是一个 (N,5) 浮点数组。

当col[0] 是[1, 3] 时，mydata[:, col[i]] 将是 (N,2)；对于其他i，它可以是 (N,0) 或 (N,1), (N,3)。

b[i, col[i]] 是 (2,)（或 (0,),(1,),(3,)）。形状有相当明显的不匹配。不能将位 (N,2) 数组放入 (2,) 槽中。

你为什么要这样构造一个数组？ None、数字和具有 (N,1) 和 (N,2) 等形状的数组的混合？

---

我认为你需要添加一个迭代：

for j in col[i]:
    b[i, j] = mydata[:, j]

这应该将 (N,) 数组分配给 b 的 b[i,j] 元素。