mnist_gan.m

trainingData = <|
   "random" -> RandomInteger[{1, 10}, Length[mnistDigits]], 
   "Input" -> 
    Map[RandomReal[{-0.05, 0.05}, {1, 28, 28}] + 
       ArrayReshape[ImageData[#], {1, 28, 28}] &, mnistDigits]|>;
generator = 
  NetChain[{EmbeddingLayer[8*6*6, 10], ReshapeLayer[{8, 6, 6}], 
    DeconvolutionLayer[8, 4, "Stride" -> 2], Ramp, 
    DeconvolutionLayer[1, 4, "Stride" -> 2, "PaddingSize" -> 1], 
    LogisticSigmoid}];
discriminator = 
  NetChain[{ConvolutionLayer[4, 4], Tanh, PoolingLayer[3, 1], 16, 
    Ramp, 1}, "Input" -> {1, 28, 28}];
wganNet = 
  NetGraph[<|"gen" -> generator, 
    "discrimop" -> NetMapOperator[discriminator], 
    "cat" -> CatenateLayer[], 
    "reshape" -> ReshapeLayer[{2, 1, 28, 28}], 
    "flat" -> FlattenLayer[], "total" -> SummationLayer[], 
    "scale" -> 
     ConstantTimesLayer["Scaling" -> {-1, 1}]|>, {NetPort["random"] ->
      "gen" -> "cat", NetPort["Input"] -> "cat", 
    "cat" -> 
     "reshape" -> "discrimop" -> "flat" -> "scale" -> "total"}, 
   "Input" -> {1, 28, 28}];
NetTrain[wganNet, trainingData, "Output", 
  Method -> {"ADAM", "Beta1" -> 0.5, "LearningRate" -> 0.01, 
    "WeightClipping" -> {{"discrimop", 1} -> 1, 
      "discrimop" -> 0.001}}, 
  TrainingProgressReporting -> {progressFuncCreator[Range[10]], 
    "Interval" -> Quantity[0.3, "Seconds"]}, 
  LearningRateMultipliers -> {"scale" -> 0, "gen" -> -0.05}, 
  TargetDevice -> "GPU", BatchSize -> 32, MaxTrainingRounds -> 5000];

Thank you for sharing! In v12, there should be a change to make this code run: "WeightClipping" -> {"discrimop" -> 0.001}.

Thank you for sharing!
The following error appears in Mathematica 11.3.0.0: "FlattenLayer::nettypeinc: Type inconsistency in FlattenLayer: dynamic dimensions cannot be flattened with other levels."
I would be very grateful if someone could help me.