How is the number of weights in the first layer of fitcnet determined?

I was looking at the ionosphere example in the fitcnet documentation. I can specify the number of neurons in the fully connected layers, say 25 and 20, and LayerWeights{2} yields 20x25, which makes sense. However, LayerWeights{1} is 25x34, and I have no idea where the 34 comes from. The second dimensions stays 34 even when the middle layers are changed. So why 34?

Answer 1

[Taylor]

The matrix "X" in the ionosphere dataset contains 34 features. Your input layer must always have the same number of columns as features in your data.