How is this not allowed in Numpy?

Posted by Delicious-Turnip8373@reddit | learnprogramming | View on Reddit | 6 comments

arr = np.array([
                [[1, 2, 3, 3], [1, 0, 0]],
                [[1, 2, 3, 3], [1, 2, 0]],
                [[1, 3, 4, 3], [1, 3, 4]]
                ])

This gives error:
ValueError: setting an array element with a sequence. The requested array has an inhomogeneous shape after 2 dimensions. The detected shape was (3, 2) + inhomogeneous part.

How is this not allowed? It is an array where each element is an array with 2 elements, the first being an array of length 4 and the second being an arra of length 3.

This is ridiculuous! How am I supposed to do anything with tensorflow like this. 
Using Numpy version 1.26.4

[-]

dmazzoni@reddit

Because np.array isn't just a bunch of nested arrays, it's specifically a single multidimensional array, and intuitively what the error message is telling you is that it has to be "rectangular".

Let's take a simpler example of a matrix.

This is a matrix:

[ 1 2 3 ]
[ 4 5 6 ]

But this is not a matrix:

[ 1 2 3 ]
[ 4 5 ]

But that's what you're trying to do (just in 3-dimensions).

A matrix must be rectangular.

A 3-d tensor, or a 3-d numpy array, must also be "rectangular" in a 3-d sense. Not sure the more precise technical word, but the length of every dimension must be consistent. Without that property, you couldn't do any tensor math.

This is ridiculuous! How am I supposed to do anything with tensorflow like this.

I think you're misunderstanding how to use tensorflow or math or both. Can you explain more about what you're trying to do?

[-]

Educational_Map_1369@reddit

Rectangular in 3d = cubical?

[-]

dmazzoni@reddit

Not quite right, because a cube is the 3d version of a square, where all sides are equal. I think you can call it a cuboid or rectangular prism, but those aren't common enough words if you don't work in that field.

[-]

Morsie@reddit

Take a look at the documentation https://numpy.org/doc/stable/reference/arrays.ndarray.html

An ndarray is a (usually fixed-size) multidimensional container of items of the same type and size. The number of dimensions and items in an array is defined by its shape, which is a tuple of N non-negative integers that specify the sizes of each dimension.

What should the shape of your array be (3, 2, (4 or 3?))? At each level your array needs to have the same size.

Making a simpler example:

arr = np.array([
                [1,2,3],
                [4,5]
               ])

You can see cleary that shape of the array is not clearly defined. The general definition of the arrys in numpy is here https://numpy.org/doc/stable/reference/arrays.html

[-]

CatSwolo@reddit

Can you be specific with what you exactly want? As of now, you are intending to make a out of shape 3 dimensional array. But seems like you are intending to make 3 matrices, right?

If that is the case then you gotta do this:

arr1 = np.array([
    [1, 2, 3, 3],
    [1, 2, 3, 3],
    [1, 2, 3, 3],
])

arr2 = np.array([
    [1, 0, 0],
    [1, 2, 0],
    [1, 3, 4],
])

Either way, the problem with your code is that 2 dimensional matrices have to be a MxN shape. You cant make a row with 4 elements and the following row has only 3 rows. Simply said a 2 dimensional matrix should always be in a form of a rectangle, not a trapezoid.

[-]

BadBoyJH@reddit

How is this not allowed? It is an array where each element is an array with 2 elements

It isn't. It's a 3D array of integers, and the arrays aren't the same size, which they need to be in numpy.