将（n，m）形numpy数组转换为（n，m，1,1）形数组

My starting point is a pandas data frame which I convert into a numpy array:

我的出发点是一个pandas数据框,我将其转换为numpy数组:

> df = pd.DataFrame({"a":[1,2,3,4],"b":[4,5,6,7],"c":[7,8,9,10]})
> arr = df.as_matrix()

The array is now 2-dimensional of shape (4,3):

该阵列现在是二维形状(4,3):

> arr
array([[ 1,  4,  7],
       [ 2,  5,  8],
       [ 3,  6,  9],
       [ 4,  7, 10]])

What I would like to do is to convert arr into its 4-dimensional and (4,3,1,1) shaped equivalent by effectively mapping every singular element like f.x. 5 onto [[5]].

我想做的是通过有效地映射每个奇异元素,如f.x,将arr转换为其4维和(4,3,1,1)形状的等价物。 5到[[5]]。

The new arr would be:

新的arr将是:

array([[ [[1]],  [[4]],  [[7]]  ],
       [ [[2]],  [[5]],  [[8]]  ],
       [ [[3]],  [[6]],  [[9]]  ],
       [ [[4]],  [[7]],  [[10]] ]])

How would I do that elegantly and fast?

我怎么能优雅而快速地做到这一点?

1 个解决方案

#1

Do arr[:, :, None, None] to add two extra axes. Here is an example:

arr [:,:,None,None]添加两个额外的轴。这是一个例子:

In [5]: arr[:, :, None, None].shape
Out[5]: (4, 3, 1, 1)

None in indexing is a synonym for np.newaxis, which selects data and adds a new axis. Many people would prefer to write the above as

索引中的无是np.newaxis的同义词,它选择数据并添加新轴。很多人宁愿把上面的内容写成

arr[:, :, np.newaxis, np.newaxis]

arr [:,:,np.newaxis,np.newaxis]

for legibility reasons

出于易读的原因

#1