np.squeeze what means axis=-1?

Question

The numpy.squeeze function takes an axis parameter which should be an integer. Sometimes I see the axis parameter being set to axis=-1, however the documentation does not explain what a negative integer does in this case.

What does numpy.squeeze do exactly when axis=-1 is set?

Answer 1

According to the docs, the axis parameter

Selects a subset of the single-dimensional entries in the shape.

Basically you choose along which axis you want to apply squeeze.

Eg

a = numpy.ones((1,2,2,1))
a.shape

(1,2,2,1)

numpy.squeeze(a,axis=0).shape

(2,2,1)

numpy.squeeze(a,axis=-1).shape

(1,2,2)

So axis=-1 chooses the last axis by numpy convention

If the size of the last axis is more than one, it raises an error.

Answer 2

According to documentation, squeeze simply removes axes of length one from the array.

you can specify the axis to remove:

x = np.array([[[0], [1], [2]]])
x.shape
(1, 3, 1)

np.squeeze(x, axis=0).shape
(3, 1)

or not, where it will remove all axes with length One:

np.squeeze(x).shape
(3,)