Sometimes you may need to transform higher-dimensional arrays into simpler 1D arrays. This is where array flattening comes into play.

This operation is useful when you need to process the elements of an array one by one or when you want to make data more suitable for certain algorithms.

There are three possible options for flattening in NumPy:

• Using the ndarray.reshape(-1) method or the numpy.reshape(array, -1) function;
• Using the ndarray.ravel() method or the numpy.ravel(array) function;
• Using the ndarray.flatten() method.

reshape(-1)

The .reshape(-1) method or the reshape(array, -1) function will return a contiguous flattened array with the same number of elements.

As we have already mentioned in the previous chapter, -1 automatically calculates the size of the dimension based on the original array's size. Since we pass only a single integer for shape, a 1D array with the same number of elements is returned.

Here is an example:

As you can see, everything is simple here. However, the .reshape() method or the respective function returns a view of the original array, so any changes made to the reshaped array will also affect the original array.

Using flattened_array = np.reshape(array_2d, -1) can be used instead of calling the method.

ravel()

The ndarray.ravel() method or the numpy.ravel(array) function works the same as reshape(-1) and also returns a view of the original array:

flattened_array = np.ravel(array_2d) can be used instead of calling the method.

ndarray.flatten()

In case you want a copy of the original array, not a view, you can use the .flatten() method:

Now the changes in the flattened array do not affect the original array.

Note

You can always copy a view of an array to create a separate object and modify this copy without affecting the original array.