Am I the only one doing "+1" to the number of output units?

Course Q&A TensorFlow Developer Professional Convolutional Neural Networks in TensorFlow
1

For the final dense layer of my model I’m doing units=len(np.unique(training_labels))+1 (if I don’t put the +1 I get a nan loss and very low accuracy (in the region of 0.04). Why does the model need the +1? Is it because there’s a gap in the labels? I noticed the number of unique values “skips” number 9 (see below)
np.unique(training_labels) yields array([ 0., 1., 2., 3., 4., 5., 6., 7., 8., 10., 11., 12., 13., 14., 15., 16., 17., 18., 19., 20., 21., 22., 23., 24.]), which has a length of 24

Thank you for your thoughts on this

2

For an input image, the last layer predicts the likelihood of assignment to a particular class.
The maximum number we have for a class label is 24 in this case. So, the last layer needs to have 25 units.

3

so it assumes the label encodings are without gaps, right?

4

That is correct. Use np.max

1 Like
5

Thank you for the explanation