Numpy array dimensions

Unwrapping the .shape gift

.shape explained

The .shape attribute provides a tuple of integers, each digit telling the size of the corresponding dimension (axis) in the NumPy array. This tuple's length is the rank of the array.

weird_array = np.array([[[1], [2], [3]], [[4], [5], [6]]])
print(weird_array.shape)  # Output: (2, 3, 1)


Dimension indexing

To access dimensions individually, you can index the shape tuple:

print(my_array.shape[0])  # Number of sweet rows 
print(my_array.shape[1])  # Number of delicious columns


Checking dimensions with .ndim

Need the count of dimensions (axes) in the array? There's an attribute for that! .ndim:

print(weird_array.ndim)  # Output: 3


Where shape matters

Shape in broadcasting

Understanding the shape is a must when you're about to perform operations like broadcasting. It ensures dimensions align correctly, preventing nasty surprises!

candy_1 = np.array([1, 2, 3])
candy_2 = np.array([[4], [5], [6]])
print((candy_1 + candy_2).shape)  # Broadcasting sugars (3,) and (3,1)


Reshaping arrays for fun

Feeling like a potter shaping clay? .reshape method alters the dimensions and your array keeps data intact. Perfect for pre-shaping data for machine learning models.

reshaped_sweets = my_array.reshape((3, 2))
print(reshaped_sweets.shape)  # Output: (3, 2)


Iterating over multidimensional arrays

With nested loops over arrays, knowing the size of each dimension can save you from the "notorious" IndexError and bring about efficient candy distribution.

for i in range(my_array.shape[0]):
    for j in range(my_array.shape[1]):
        print(f"{my_array[i, j]} candy, please!", end=' ')
    print()


Shape-shifting tricks with arrays

Flattening

You can unroll your array to 1D using the .flatten() or .ravel() methods. These methods are sweet when dealing with functions demanding one-dimensional inputs!

flat_candies = weird_array.flatten()
print(flat_candies.shape)  # Output: (6,)


Pumping up dimensions

Need more dimensions? Use np.newaxis or .reshape to add an axis. Mostly handy in deep learning convolution!

pumped_candies = candy_1[:, np.newaxis]
print(pumped_candies.shape)  # Output: (3, 1)


Slicing and indexing

Shape can be affected by array slicing and advanced indexing. Slicing keeps the dimension, while integer indexing reduces it:

sliced_candies = weird_array[:, :2, :]
print(sliced_candies.shape)  # Output: (2, 2, 1)

indexed_candies = weird_array[0, :, :]
print(indexed_candies.shape)  # Output: (3, 1)
