Expected shapes


#1

Hi, I have training data where the inputs are both numeric (4 of these) and categorical (577 categories), and the output is categorical (807 of these). No matter what model I create (and it always has the green tick to show it’s valid), I always get the error message (when training):

Error when checking model target: expected dense_xx to have shape(None,1) but got array with shape(1,807).

I get this even if I use the AutoML tool. I get this even if all I have is a single dense layer between input and output. I don’t understand - please help!

More generally, I love the software but have no idea what most of it means - is there no documentation, other than a few silent videos of specific examples?

Thanks!
Phil


#2

@phildrew yes, documentation is one area where we need to prioritize. Currently there is no documentation.

That said, we can help you to resolve this issue. Is it possible to share the project config (exported YAML file from model tab)?

What is your loss function ? is it binary_crossentropy or categorical_crossentropy?