On Formatting Input Data When Training Neural Network in Mathematica

Question

I'm trying to train the following structure of neural networknet
with customized loss functionlossnet

each code and structure is below.

net = NetChain[{LinearLayer[10], BatchNormalizationLayer[], 
   ElementwiseLayer[Tanh], LinearLayer[10]}, "Input" -> 10, 
  "Output" -> 10]

lossnet = NetGraph[<|
   "rest" -> SequenceRestLayer[],
   "reverse" -> SequenceReverseLayer[],
   "firstelement" -> SequenceLastLayer[],
   "elementwise" -> ElementwiseLayer[Power[# - 1]^2 &],
   "most" -> SequenceMostLayer[],
   "last" -> ThreadingLayer[(#1 - #2 - 1)^2 &],
   "out" -> SummationLayer[],
   "summary" -> ThreadingLayer[Plus]
   |>,
  {
   NetPort["Input"] -> "rest",
   NetPort["Input"] -> "reverse",
   "reverse" -> "firstelement",
   "firstelement" -> "elementwise",
   NetPort["Input"] -> "most",
   "rest" -> "last",
   "most" -> "last",
   "last" -> "out",
   {"out", "elementwise"} -> "summary"}]

but when start training,I encountered this error.

NetTrain[net, <|NetPort["Input"] -> {Range[1, 10]}|>, 
 LossFunction -> lossnet]

NetTrain::missinslot: Training data specification should include values for the port Input, but only contains values for the port NetPort[Input]. You may need to explicitly specify the loss port(s) of the net using LossFunction.

it seems that something related to my input is wrong,but I don't know where it is.
perhaps,this is a rudimentary mistake,though.

what is wrong?

Answer 1

For most neural networks, you need to specify output data as well as input data for NetTrain. Your code only specifies input data.

NetTrain[net, <|"Input" -> {Range[1, 10]}, "Output" -> {ConstantArray[0, {10}]}|>,
   LossFunction -> lossnet]

works, as will the shorthand {examplein1 -> exampleout1, examplein2 -> exampleout2, ...} like:

NetTrain[net, {examplein1 -> exampleout1, examplein2 -> exampleout2, ...}, 
   LossFunction -> lossnet]