Making sense of tf.argmax for tensors with more than one axis

[MithilRocks]

During the lecture for finding positional minimum and maximum of a tensor, I came across this in the documentation of Tensorflow:

B = tf.constant([[2, 20, 30, 3, 6], [3, 11, 16, 1, 8],
                 [14, 45, 23, 5, 27]])
tf.math.argmax(B)
<tf.Tensor: shape=(5,), dtype=int64, numpy=array([0, 0, 0, 1, 0])>

I am unable to wrap my head around why the output is [0,0,0,1,0]. Any explanation?

[gauravreddy08]

Hey @MithilRocks ,

I don't know which documentation you're referring to, but the Tensorflow argmax Documentation I found looks good. Check the link.

Here's the code I tried out in the Notebook.