Simplified cifar10 architecture

src/test/java/de/monticore/lang/monticar/emadl/IntegrationCaffe2Test.java

@@ -74,7 +74,7 @@ public class IntegrationCaffe2Test extends AbstractSymtabTest {
     public void testDontRetrain1() {
         // The training hash is stored during the first training, so the second one is skipped
         Log.getFindings().clear();
-        String[] args = {"-m", "src/test/resources/models/", "-r", "cifar10.Cifar10Classifier", "-b", "CAFFE2"};
+        String[] args = {"-m", "src/test/resources/models/", "-r", "simpleCifar10.Cifar10Classifier", "-b", "CAFFE2"};
         EMADLGeneratorCli.main(args);
         assertTrue(Log.getFindings().isEmpty());
         

src/test/resources/models/data_paths.txt

 
 cifar10.CifarNetwork src/test/resources/training_data
+simpleCifar10.CifarNetwork src/test/resources/training_data
 cNNCalculator.Network src/test/resources/training_data
 MultipleOutputs data/MultipleOutputs
 InstanceTest.NetworkB data/InstanceTest.NetworkB

src/test/resources/models/simpleCifar10/ArgMax.emadl

+package simpleCifar10;
+
+component ArgMax<Z(1:oo) n = 2>{
+    ports in Q^{n} inputVector,
+          out Z(0:oo) maxIndex;
+
+    implementation Math{
+         maxIndex = 0;
+         Q maxValue = inputVector(1);
+
+         for i = 2:n
+            if inputVector(i) > maxValue
+                maxIndex = i - 1;
+                maxValue = inputVector(i);
+            end
+         end
+    }
+
+}
\ No newline at end of file

src/test/resources/models/simpleCifar10/Cifar10Classifier.emadl

+package simpleCifar10;
+import Network;
+import CalculateClass;
+
+component Cifar10Classifier{
+    ports in Z(0:255)^{3, 32, 32} image,
+         out Z(0:9) classIndex;
+
+    instance CifarNetwork<10> net;
+
+    instance ArgMax<10> calculateClass;
+
+    connect image -> net.data;
+    connect net.softmax -> calculateClass.inputVector;
+    connect calculateClass.maxIndex -> classIndex;
+}
\ No newline at end of file

src/test/resources/models/simpleCifar10/CifarNetwork.cnnt

+configuration CifarNetwork{
+     num_epoch:10
+     batch_size:5
+     normalize:true
+     context:cpu
+     load_checkpoint:false
+     optimizer:adam{
+         learning_rate:0.01
+         learning_rate_decay:0.8
+         step_size:1000
+         weight_decay:0.0001
+     }
+}

src/test/resources/models/simpleCifar10/CifarNetwork.emadl

+package simpleCifar10;
+
+component CifarNetwork<Z(2:oo) classes = 10>{
+    ports in Z(0:255)^{3, 32, 32} data,
+         out Q(0:1)^{classes} softmax;
+
+    implementation CNN {
+        def conv(kernel, channels, stride=1, act=true){
+            Convolution(kernel=(kernel,kernel), channels=channels, stride=(stride,stride)) ->
+        }
+        def resLayer(channels, stride=1, addSkipConv=false){
+            (
+                conv(kernel=3, channels=channels, stride=stride) ->
+                conv(kernel=3, channels=channels, act=false)
+            |
+                conv(kernel=1, channels=channels, stride=stride, act=false, ? = addSkipConv)
+            ) ->
+            Add() ->
+        }
+
+        data ->
+        resLayer(channels=8, addSkipConv=true) ->
+        resLayer(channels=16, stride=2, addSkipConv=true) ->
+        resLayer(channels=16, ->=2) ->
+        resLayer(channels=32, stride=2, addSkipConv=true) ->
+        resLayer(channels=32, ->=2) ->
+        resLayer(channels=64, stride=2, addSkipConv=true) ->
+        resLayer(channels=64, ->=2) ->
+        GlobalPooling(pool_type="avg") ->
+        FullyConnected(units=128) ->
+        Dropout()->
+        FullyConnected(units=classes) ->
+        Softmax() ->
+        softmax
+
+    }
+}
\ No newline at end of file