Merge rnn into develop

pom.xml

@@ -15,9 +15,9 @@
   <properties>
 
     <!-- .. SE-Libraries .................................................. -->
-      <CNNArch.version>0.3.0-SNAPSHOT</CNNArch.version>
+      <CNNArch.version>0.3.1-SNAPSHOT</CNNArch.version>
       <CNNTrain.version>0.3.2-SNAPSHOT</CNNTrain.version>
-      <CNNArch2MXNet.version>0.2.15-SNAPSHOT</CNNArch2MXNet.version>
+      <CNNArch2MXNet.version>0.2.16-SNAPSHOT</CNNArch2MXNet.version>
       <embedded-montiarc-math-opt-generator>0.1.4</embedded-montiarc-math-opt-generator>
       <EMADL2PythonWrapper.version>0.0.1</EMADL2PythonWrapper.version>
     
@@ -116,6 +116,12 @@
             <scope>test</scope>
         </dependency>
 
+        <dependency>
+            <groupId>com.github.stefanbirkner</groupId>
+            <artifactId>system-rules</artifactId>
+            <version>1.3.0</version>
+        </dependency>
+
         <dependency>
             <groupId>org.mockito</groupId>
             <artifactId>mockito-core</artifactId>

src/main/java/de/monticore/lang/monticar/cnnarch/gluongenerator/CNNArch2Gluon.java

@@ -33,11 +33,15 @@ import java.util.Map;
 
 public class CNNArch2Gluon extends CNNArch2MxNet {
 
+    public CNNArch2Gluon() {
+        architectureSupportChecker = new CNNArch2GluonArchitectureSupportChecker();
+        layerSupportChecker = new CNNArch2GluonLayerSupportChecker();
+    }
+
     //check cocos with CNNArchCocos.checkAll(architecture) before calling this method.
     @Override
     public Map<String, String> generateStrings(ArchitectureSymbol architecture){
         Map<String, String> fileContentMap = compileFileContentMap(architecture);
-        checkValidGeneration(architecture);
         return fileContentMap;
     }
 
@@ -85,6 +89,9 @@ public class CNNArch2Gluon extends CNNArch2MxNet {
         temp = controller.process("CNNPredictor", Target.CPP);
         fileContentMap.put(temp.getKey(), temp.getValue());
 
+        temp = controller.process("CNNSupervisedTrainer", Target.PYTHON);
+        fileContentMap.put(temp.getKey(), temp.getValue());
+
         temp = controller.process("execute", Target.CPP);
         fileContentMap.put(temp.getKey().replace(".h", ""), temp.getValue());
 

src/main/java/de/monticore/lang/monticar/cnnarch/gluongenerator/CNNArch2GluonArchitectureSupportChecker.java

+package de.monticore.lang.monticar.cnnarch.gluongenerator;
+
+import de.monticore.lang.monticar.cnnarch._symboltable.ArchitectureSymbol;
+import de.monticore.lang.monticar.cnnarch.mxnetgenerator.ArchitectureSupportChecker;
+
+public class CNNArch2GluonArchitectureSupportChecker extends ArchitectureSupportChecker {
+
+    public CNNArch2GluonArchitectureSupportChecker() {}
+
+    /*protected boolean checkMultipleStreams(ArchitectureSymbol architecture) {
+        return true;
+    }*/
+
+    protected boolean checkMultipleInputs(ArchitectureSymbol architecture) {
+        return true;
+    }
+
+    protected boolean checkMultipleOutputs(ArchitectureSymbol architecture) {
+        return true;
+    }
+
+}

src/main/java/de/monticore/lang/monticar/cnnarch/gluongenerator/CNNArch2GluonLayerSupportChecker.java

+package de.monticore.lang.monticar.cnnarch.gluongenerator;
+
+import de.monticore.lang.monticar.cnnarch.predefined.AllPredefinedLayers;
+import de.monticore.lang.monticar.cnnarch.mxnetgenerator.LayerSupportChecker;
+
+public class CNNArch2GluonLayerSupportChecker extends LayerSupportChecker {
+
+    public CNNArch2GluonLayerSupportChecker() {
+        supportedLayerList.add(AllPredefinedLayers.FULLY_CONNECTED_NAME);
+        supportedLayerList.add(AllPredefinedLayers.CONVOLUTION_NAME);
+        supportedLayerList.add(AllPredefinedLayers.SOFTMAX_NAME);
+        supportedLayerList.add(AllPredefinedLayers.SIGMOID_NAME);
+        supportedLayerList.add(AllPredefinedLayers.TANH_NAME);
+        supportedLayerList.add(AllPredefinedLayers.RELU_NAME);
+        supportedLayerList.add(AllPredefinedLayers.DROPOUT_NAME);
+        supportedLayerList.add(AllPredefinedLayers.POOLING_NAME);
+        supportedLayerList.add(AllPredefinedLayers.GLOBAL_POOLING_NAME);
+        supportedLayerList.add(AllPredefinedLayers.LRN_NAME);
+        supportedLayerList.add(AllPredefinedLayers.BATCHNORM_NAME);
+        supportedLayerList.add(AllPredefinedLayers.SPLIT_NAME);
+        supportedLayerList.add(AllPredefinedLayers.GET_NAME);
+        supportedLayerList.add(AllPredefinedLayers.ADD_NAME);
+        supportedLayerList.add(AllPredefinedLayers.CONCATENATE_NAME);
+        supportedLayerList.add(AllPredefinedLayers.FLATTEN_NAME);
+        supportedLayerList.add(AllPredefinedLayers.ONE_HOT_NAME);
+    }
+
+}

src/main/java/de/monticore/lang/monticar/cnnarch/gluongenerator/CNNArch2GluonTemplateController.java

@@ -71,7 +71,7 @@ public class CNNArch2GluonTemplateController extends CNNArchTemplateController {
 
         if (layer.isAtomic()){
             ArchitectureElementSymbol nextElement = layer.getOutputElement().get();
-            if (!isSoftmaxOutput(nextElement) && !isLogisticRegressionOutput(nextElement)){
+            if (!isSoftmaxOutput(nextElement) && !isLogisticRegressionOutput(nextElement) && !isOneHotOutput(nextElement)){
                 String templateName = layer.getDeclaration().getName();
                 include(TEMPLATE_ELEMENTS_DIR_PATH, templateName, writer, netDefinitionMode);
             }

src/main/java/de/monticore/lang/monticar/cnnarch/gluongenerator/CNNTrain2Gluon.java

@@ -114,9 +114,6 @@ public class CNNTrain2Gluon extends CNNTrain2MxNet {
         if (configData.isSupervisedLearning()) {
             String cnnTrainTemplateContent = templateConfiguration.processTemplate(ftlContext, "CNNTrainer.ftl");
             fileContentMap.put("CNNTrainer_" + getInstanceName() + ".py", cnnTrainTemplateContent);
-
-            String cnnSupervisedTrainerContent = templateConfiguration.processTemplate(ftlContext, "CNNSupervisedTrainer.ftl");
-            fileContentMap.put("supervised_trainer.py", cnnSupervisedTrainerContent);
         } else if (configData.isReinforcementLearning()) {
             final String trainerName = "CNNTrainer_" + getInstanceName();
             final RLAlgorithm rlAlgorithm = configData.getRlAlgorithm();

src/main/java/de/monticore/lang/monticar/cnnarch/gluongenerator/reinforcement/critic/CriticNetworkGenerator.java

@@ -3,9 +3,10 @@ package de.monticore.lang.monticar.cnnarch.gluongenerator.reinforcement.critic;
 import com.google.common.collect.Lists;
 import de.monticore.lang.monticar.cnnarch._symboltable.*;
 import de.monticore.lang.monticar.cnnarch.gluongenerator.CNNArch2Gluon;
+import de.monticore.lang.monticar.cnnarch.gluongenerator.CNNArch2GluonArchitectureSupportChecker;
+import de.monticore.lang.monticar.cnnarch.gluongenerator.CNNArch2GluonLayerSupportChecker;
 import de.monticore.lang.monticar.cnnarch.mxnetgenerator.CNNArchSymbolCompiler;
 import de.monticore.lang.monticar.cnnarch.mxnetgenerator.TemplateConfiguration;
-import de.monticore.lang.monticar.cnnarch.mxnetgenerator.checker.AllowAllLayerSupportChecker;
 import de.monticore.lang.monticar.cnntrain._symboltable.ConfigurationSymbol;
 import de.monticore.lang.monticar.cnntrain.annotations.Range;
 import de.monticore.lang.monticar.cnntrain.annotations.TrainedArchitecture;
@@ -81,7 +82,8 @@ public class CriticNetworkGenerator {
         gluonGenerator.setGenerationTargetPath(this.getGenerationTargetPath());
 
         Map<String, String> fileContentMap = new HashMap<>();
-        CNNArchSymbolCompiler symbolCompiler = new CNNArchSymbolCompiler(new AllowAllLayerSupportChecker());
+        CNNArchSymbolCompiler symbolCompiler = new CNNArchSymbolCompiler(new CNNArch2GluonArchitectureSupportChecker(),
+                                                                         new CNNArch2GluonLayerSupportChecker());
         ArchitectureSymbol architectureSymbol = symbolCompiler.compileArchitectureSymbolFromModelsDir(directoryOfCnnArchFile, criticNetworkName);
         architectureSymbol.setComponentName(criticNetworkName);
         fileContentMap.putAll(gluonGenerator.generateStringsAllowMultipleIO(architectureSymbol, true));

src/main/resources/templates/gluon/CNNCreator.ftl

@@ -6,7 +6,7 @@ from CNNNet_${tc.fullArchitectureName} import Net
 class ${tc.fileNameWithoutEnding}:
     _model_dir_ = "model/${tc.componentName}/"
     _model_prefix_ = "model"
-    _input_shapes_ = [<#list tc.architecture.inputs as input>(${tc.join(input.definition.type.dimensions, ",")},)<#if input?has_next>,</#if></#list>]
+    _input_shapes_ = [<#list tc.architecture.inputs as input>(${tc.join(input.definition.type.dimensions, ",")},)<#sep>, </#list>]
 
     def __init__(self):
         self.weight_initializer = mx.init.Normal()
@@ -43,12 +43,11 @@ class ${tc.fileNameWithoutEnding}:
             self.net.load_parameters(self._model_dir_ + param_file)
             return lastEpoch
 
-
     def construct(self, context, data_mean=None, data_std=None):
         self.net = Net(data_mean=data_mean, data_std=data_std)
         self.net.collect_params().initialize(self.weight_initializer, ctx=context)
         self.net.hybridize()
-        self.net(<#list tc.architecture.inputs as input>mx.nd.zeros((1,)+self._input_shapes_[${input?index}], ctx=context)<#if input?has_next>,</#if></#list>)
+        self.net(<#list tc.architecture.inputs as input>mx.nd.zeros((1,) + self._input_shapes_[${input?index}], ctx=context)<#sep>, </#list>)
 
         if not os.path.exists(self._model_dir_):
             os.makedirs(self._model_dir_)

src/main/resources/templates/gluon/CNNDataLoader.ftl

@@ -3,8 +3,9 @@ import h5py
 import mxnet as mx
 import logging
 import sys
+from mxnet import nd
 
-class ${tc.fullArchitectureName}DataLoader:
+class ${tc.fileNameWithoutEnding}:
     _input_names_ = [${tc.join(tc.architectureInputs, ",", "'", "'")}]
     _output_names_ = [${tc.join(tc.architectureOutputs, ",", "'", "_label'")}]
 
@@ -14,21 +15,38 @@ class ${tc.fullArchitectureName}DataLoader:
     def load_data(self, batch_size):
         train_h5, test_h5 = self.load_h5_files()
 
-        data_mean = train_h5[self._input_names_[0]][:].mean(axis=0)
-        data_std = train_h5[self._input_names_[0]][:].std(axis=0) + 1e-5
+        train_data = {}
+        data_mean = {}
+        data_std = {}
+
+        for input_name in self._input_names_:
+            train_data[input_name] = train_h5[input_name]
+            data_mean[input_name] = nd.array(train_h5[input_name][:].mean(axis=0))
+            data_std[input_name] = nd.array(train_h5[input_name][:].std(axis=0) + 1e-5)
+
+        train_label = {}
+        for output_name in self._output_names_:
+            train_label[output_name] = train_h5[output_name]
+
+        train_iter = mx.io.NDArrayIter(data=train_data,
+                                       label=train_label,
+                                       batch_size=batch_size)
 
-        train_iter = mx.io.NDArrayIter(train_h5[self._input_names_[0]],
-                                       train_h5[self._output_names_[0]],
-                                       batch_size=batch_size,
-                                       data_name=self._input_names_[0],
-                                       label_name=self._output_names_[0])
         test_iter = None
+
         if test_h5 != None:
-            test_iter = mx.io.NDArrayIter(test_h5[self._input_names_[0]],
-                                          test_h5[self._output_names_[0]],
-                                          batch_size=batch_size,
-                                          data_name=self._input_names_[0],
-                                          label_name=self._output_names_[0])
+            test_data = {}
+            for input_name in self._input_names_:
+                test_data[input_name] = test_h5[input_name]
+
+            test_label = {}
+            for output_name in self._output_names_:
+                test_label[output_name] = test_h5[output_name]
+
+            test_iter = mx.io.NDArrayIter(data=test_data,
+                                          label=test_label,
+                                          batch_size=batch_size)
+
         return train_iter, test_iter, data_mean, data_std
 
     def load_h5_files(self):
@@ -36,21 +54,39 @@ class ${tc.fullArchitectureName}DataLoader:
         test_h5 = None
         train_path = self._data_dir + "train.h5"
         test_path = self._data_dir + "test.h5"
+
         if os.path.isfile(train_path):
             train_h5 = h5py.File(train_path, 'r')
-            if not (self._input_names_[0] in train_h5 and self._output_names_[0] in train_h5):
-                logging.error("The HDF5 file '" + os.path.abspath(train_path) + "' has to contain the datasets: "
-                              + "'" + self._input_names_[0] + "', '" + self._output_names_[0] + "'")
-                sys.exit(1)
-            test_iter = None
+
+            for input_name in self._input_names_:
+                if not input_name in train_h5:
+                    logging.error("The HDF5 file '" + os.path.abspath(train_path) + "' has to contain the dataset "
+                                  + "'" + input_name + "'")
+                    sys.exit(1)
+
+            for output_name in self._output_names_:
+                if not output_name in train_h5:
+                    logging.error("The HDF5 file '" + os.path.abspath(train_path) + "' has to contain the dataset "
+                                  + "'" + output_name + "'")
+                    sys.exit(1)
+
             if os.path.isfile(test_path):
                 test_h5 = h5py.File(test_path, 'r')
-                if not (self._input_names_[0] in test_h5 and self._output_names_[0] in test_h5):
-                    logging.error("The HDF5 file '" + os.path.abspath(test_path) + "' has to contain the datasets: "
-                                  + "'" + self._input_names_[0] + "', '" + self._output_names_[0] + "'")
-                    sys.exit(1)
+
+                for input_name in self._input_names_:
+                    if not input_name in test_h5:
+                        logging.error("The HDF5 file '" + os.path.abspath(test_path) + "' has to contain the dataset "
+                                      + "'" + input_name + "'")
+                        sys.exit(1)
+
+                for output_name in self._output_names_:
+                    if not output_name in test_h5:
+                        logging.error("The HDF5 file '" + os.path.abspath(test_path) + "' has to contain the dataset "
+                                      + "'" + output_name + "'")
+                        sys.exit(1)
             else:
                 logging.warning("Couldn't load test set. File '" + os.path.abspath(test_path) + "' does not exist.")
+
             return train_h5, test_h5
         else:
             logging.error("Data loading failure. File '" + os.path.abspath(train_path) + "' does not exist.")

src/main/resources/templates/gluon/CNNNet.ftl

@@ -2,6 +2,16 @@ import mxnet as mx
 import numpy as np
 from mxnet import gluon
 
+class OneHot(gluon.HybridBlock):
+    def __init__(self, size, **kwargs):
+        super(OneHot, self).__init__(**kwargs)
+        with self.name_scope():
+            self.size = size
+
+    def hybrid_forward(self, F, x):
+        return F.one_hot(indices=F.argmax(data=x, axis=1), depth=self.size)
+
+
 class Softmax(gluon.HybridBlock):
     def __init__(self, **kwargs):
         super(Softmax, self).__init__(**kwargs)
@@ -71,8 +81,15 @@ class NoNormalization(gluon.HybridBlock):
 class Net(gluon.HybridBlock):
     def __init__(self, data_mean=None, data_std=None, **kwargs):
         super(Net, self).__init__(**kwargs)
+        self.last_layers = {}
         with self.name_scope():
-${tc.include(tc.architecture.body, "ARCHITECTURE_DEFINITION")}
-
-    def hybrid_forward(self, F, <#list tc.architecture.inputs as input>${input}<#if input?has_next>, </#if></#list>):
-${tc.include(tc.architecture.body, "FORWARD_FUNCTION")}
\ No newline at end of file
+${tc.include(tc.architecture.streams[0], "ARCHITECTURE_DEFINITION")}
+
+    def hybrid_forward(self, F, ${tc.join(tc.architectureInputs, ", ")}):
+        <#if tc.architectureOutputs?size gt 1>
+        outputs = []
+        </#if>
+${tc.include(tc.architecture.streams[0], "FORWARD_FUNCTION")}
+        <#if tc.architectureOutputs?size gt 1>
+        return tuple(outputs)
+        </#if>

src/main/resources/templates/gluon/CNNPredictor.ftl

@@ -13,9 +13,14 @@ class ${tc.fileNameWithoutEnding}{
 public:
     const std::string json_file = "model/${tc.componentName}/model_newest-symbol.json";
     const std::string param_file = "model/${tc.componentName}/model_newest-0000.params";
-    //const std::vector<std::string> input_keys = {"data"};
-    const std::vector<std::string> input_keys = {${tc.join(tc.architectureInputs, ",", "\"", "\"")}};
-    const std::vector<std::vector<mx_uint>> input_shapes = {<#list tc.architecture.inputs as input>{1,${tc.join(input.definition.type.dimensions, ",")}}<#if input?has_next>,</#if></#list>};
+    const std::vector<std::string> input_keys = {
+<#if (tc.architectureInputs?size == 1)>
+        "data"
+<#else>
+        <#list tc.architectureInputs as inputName>"data${inputName?index}"<#sep>, </#list>
+</#if>
+    };
+    const std::vector<std::vector<mx_uint>> input_shapes = {<#list tc.architecture.inputs as input>{1, ${tc.join(input.definition.type.dimensions, ", ")}}<#sep>, </#list>};
     const bool use_gpu = false;
 
     PredictorHandle handle;
@@ -31,7 +36,11 @@ public:
     void predict(${tc.join(tc.architectureInputs, ", ", "const std::vector<float> &", "")},
                  ${tc.join(tc.architectureOutputs, ", ", "std::vector<float> &", "")}){
 <#list tc.architectureInputs as inputName>
+<#if (tc.architectureInputs?size == 1)>
         MXPredSetInput(handle, "data", ${inputName}.data(), static_cast<mx_uint>(${inputName}.size()));
+<#else>
+        MXPredSetInput(handle, "data${inputName?index}", ${inputName}.data(), static_cast<mx_uint>(${inputName}.size()));
+</#if>
 </#list>
 
         MXPredForward(handle);
@@ -71,22 +80,17 @@ public:
 
         const mx_uint num_input_nodes = input_keys.size();
 
-<#if (tc.architectureInputs?size >= 2)>
         const char* input_keys_ptr[num_input_nodes];
         for(mx_uint i = 0; i < num_input_nodes; i++){
             input_keys_ptr[i] = input_keys[i].c_str();
         }
-<#else>
-        const char* input_key[1] = { "data" };
-        const char** input_keys_ptr = input_key;
-</#if>
 
         mx_uint shape_data_size = 0;
         mx_uint input_shape_indptr[input_shapes.size() + 1];
         input_shape_indptr[0] = 0;
         for(mx_uint i = 0; i < input_shapes.size(); i++){
-            input_shape_indptr[i+1] = input_shapes[i].size();
             shape_data_size += input_shapes[i].size();
+            input_shape_indptr[i+1] = shape_data_size;
         }
 
         mx_uint input_shape_data[shape_data_size];

src/main/resources/templates/gluon/CNNSupervisedTrainer.ftl

@@ -6,7 +6,7 @@ import os
 import shutil
 from mxnet import gluon, autograd, nd
 
-class CNNSupervisedTrainer(object):
+class ${tc.fileNameWithoutEnding}:
     def __init__(self, data_loader, net_constructor, net=None):
         self._data_loader = data_loader
         self._net_creator = net_constructor
@@ -48,7 +48,7 @@ class CNNSupervisedTrainer(object):
         if self._net is None:
             if normalize:
                 self._net_creator.construct(
-                    context=mx_context, data_mean=nd.array(data_mean), data_std=nd.array(data_std))
+                    context=mx_context, data_mean=data_mean, data_std=data_std)
             else:
                 self._net_creator.construct(context=mx_context)
 
@@ -69,15 +69,18 @@ class CNNSupervisedTrainer(object):
 
         trainer = mx.gluon.Trainer(self._net.collect_params(), optimizer, optimizer_params)
 
-        if self._net.last_layer == 'softmax':
-            loss_function = mx.gluon.loss.SoftmaxCrossEntropyLoss()
-        elif self._net.last_layer == 'sigmoid':
-            loss_function = mx.gluon.loss.SigmoidBinaryCrossEntropyLoss()
-        elif self._net.last_layer == 'linear':
-            loss_function = mx.gluon.loss.L2Loss()
-        else: # TODO: Change default?
-            loss_function = mx.gluon.loss.L2Loss()
-            logging.warning("Invalid last_layer, defaulting to L2 loss")
+        loss_functions = {}
+
+        for output_name, last_layer in self._net.last_layers.items():
+            if last_layer == 'softmax':
+                loss_functions[output_name] = mx.gluon.loss.SoftmaxCrossEntropyLoss()
+            elif last_layer == 'sigmoid':
+                loss_functions[output_name] = mx.gluon.loss.SigmoidBinaryCrossEntropyLoss()
+            elif last_layer == 'linear':
+                loss_functions[output_name] = mx.gluon.loss.L2Loss()
+            else:
+                loss_functions[output_name] = mx.gluon.loss.L2Loss()
+                logging.warning("Invalid last layer, defaulting to L2 loss")
 
         speed_period = 50
         tic = None
@@ -85,11 +88,17 @@ class CNNSupervisedTrainer(object):
         for epoch in range(begin_epoch, begin_epoch + num_epoch):
             train_iter.reset()
             for batch_i, batch in enumerate(train_iter):
-                data = batch.data[0].as_in_context(mx_context)
-                label = batch.label[0].as_in_context(mx_context)
+                <#list tc.architectureInputs as input_name>
+                ${input_name}_data = batch.data[${input_name?index}].as_in_context(mx_context)
+                </#list>
+                <#list tc.architectureOutputs as output_name>
+                ${output_name}_label = batch.label[${output_name?index}].as_in_context(mx_context)
+                </#list>
+
                 with autograd.record():
-                    output = self._net(data)
-                    loss = loss_function(output, label)
+                    ${tc.join(tc.architectureOutputs, ", ", "", "_output")} = self._net(${tc.join(tc.architectureInputs, ", ", "", "_data")})
+
+                    loss = <#list tc.architectureOutputs as output_name>loss_functions['${output_name}'](${output_name}_output, ${output_name}_label)<#sep> + </#list>
 
                 loss.backward()
                 trainer.step(batch_size)
@@ -112,21 +121,41 @@ class CNNSupervisedTrainer(object):
             train_iter.reset()
             metric = mx.metric.create(eval_metric)
             for batch_i, batch in enumerate(train_iter):
-                data = batch.data[0].as_in_context(mx_context)
-                label = batch.label[0].as_in_context(mx_context)
-                output = self._net(data)
-                predictions = mx.nd.argmax(output, axis=1)
-                metric.update(preds=predictions, labels=label)
+                <#list tc.architectureInputs as input_name>
+                ${input_name}_data = batch.data[${input_name?index}].as_in_context(mx_context)
+                </#list>
+
+                labels = [
+                    <#list tc.architectureOutputs as output_name>batch.label[${output_name?index}].as_in_context(mx_context)<#sep>, </#list>
+                ]
+
+                ${tc.join(tc.architectureOutputs, ", ", "", "_output")} = self._net(${tc.join(tc.architectureInputs, ", ", "", "_data")})
+
+                predictions = [
+                    <#list tc.architectureOutputs as output_name>mx.nd.argmax(${output_name}_output, axis=1)<#sep>, </#list>
+                ]
+
+                metric.update(preds=predictions, labels=labels)
             train_metric_score = metric.get()[1]
 
             test_iter.reset()
             metric = mx.metric.create(eval_metric)
             for batch_i, batch in enumerate(test_iter):
-                data = batch.data[0].as_in_context(mx_context)
-                label = batch.label[0].as_in_context(mx_context)
-                output = self._net(data)
-                predictions = mx.nd.argmax(output, axis=1)
-                metric.update(preds=predictions, labels=label)
+                <#list tc.architectureInputs as input_name>
+                ${input_name}_data = batch.data[${input_name?index}].as_in_context(mx_context)
+                </#list>
+
+                labels = [
+                    <#list tc.architectureOutputs as output_name>batch.label[${output_name?index}].as_in_context(mx_context)<#sep>, </#list>
+                ]
+
+                ${tc.join(tc.architectureOutputs, ", ", "", "_output")} = self._net(${tc.join(tc.architectureInputs, ", ", "", "_data")})
+
+                predictions = [
+                    <#list tc.architectureOutputs as output_name>mx.nd.argmax(${output_name}_output, axis=1)<#sep>, </#list>
+                ]
+
+                metric.update(preds=predictions, labels=labels)
             test_metric_score = metric.get()[1]
 
             logging.info("Epoch[%d] Train: %f, Test: %f" % (epoch, train_metric_score, test_metric_score))

src/main/resources/templates/gluon/CNNTrainer.ftl

 import logging
 import mxnet as mx
-import supervised_trainer
 <#list configurations as config>
 import CNNCreator_${config.instanceName}
 import CNNDataLoader_${config.instanceName}
+import CNNSupervisedTrainer_${config.instanceName}
 </#list>
 
 if __name__ == "__main__":
@@ -14,9 +14,11 @@ if __name__ == "__main__":
 
 <#list configurations as config>
     ${config.instanceName}_creator = CNNCreator_${config.instanceName}.CNNCreator_${config.instanceName}()
-    ${config.instanceName}_loader = CNNDataLoader_${config.instanceName}.${config.instanceName}DataLoader()
-    ${config.instanceName}_trainer = supervised_trainer.CNNSupervisedTrainer(${config.instanceName}_loader,
-        ${config.instanceName}_creator)
+    ${config.instanceName}_loader = CNNDataLoader_${config.instanceName}.CNNDataLoader_${config.instanceName}()
+    ${config.instanceName}_trainer = CNNSupervisedTrainer_${config.instanceName}.CNNSupervisedTrainer_${config.instanceName}(
+        ${config.instanceName}_loader,
+        ${config.instanceName}_creator
+    )
 
     ${config.instanceName}_trainer.train(
 <#if (config.batchSize)??>

src/main/resources/templates/gluon/elements/Input.ftl

 <#assign mode = definition_mode.toString()>
 <#if mode == "ARCHITECTURE_DEFINITION">
-            if not data_mean is None:
-                assert(not data_std is None)
-                self.${element.name}_input_normalization = ZScoreNormalization(data_mean=data_mean, data_std=data_std)
+            if data_mean:
+                assert(data_std)
+                self.input_normalization_${element.name} = ZScoreNormalization(data_mean=data_mean['${element.name}'],
+                                                                               data_std=data_std['${element.name}'])
             else:
-                self.${element.name}_input_normalization = NoNormalization()
+                self.input_normalization_${element.name} = NoNormalization()
 
 </#if>
 <#if mode == "FORWARD_FUNCTION">
-        ${element.name} = self.${element.name}_input_normalization(${element.name})
+        ${element.name} = self.input_normalization_${element.name}(${element.name})
 </#if>
\ No newline at end of file

src/main/resources/templates/gluon/elements/OneHot.ftl

+<#assign input = element.inputs[0]>
+<#assign mode = definition_mode.toString()>
+<#assign size = element.size>
+<#if mode == "ARCHITECTURE_DEFINITION">
+            self.${element.name} = OneHot(size=${size})
+            <#include "OutputShape.ftl">
+</#if>
+<#if mode == "FORWARD_FUNCTION">
+        ${element.name} = self.${element.name}(${input})
+</#if>

src/main/resources/templates/gluon/elements/Output.ftl

 <#assign input = element.inputs[0]>
 <#assign mode = definition_mode.toString()>
 <#if mode == "ARCHITECTURE_DEFINITION">
-
     <#if element.softmaxOutput>
-        self.last_layer = 'softmax'
+        self.last_layers['${element.name}'] = 'softmax'
     <#elseif element.logisticRegressionOutput>
-        self.last_layer = 'sigmoid'
+        self.last_layers['${element.name}'] = 'sigmoid'
     <#elseif element.linearRegressionOutput>
-        self.last_layer = 'linear'
+        self.last_layers['${element.name}'] = 'linear'
+    <#elseif element.oneHotOutput>
+        self.last_layers['${element.name}'] = 'softmax'
     </#if>
 </#if>
 <#if mode == "FORWARD_FUNCTION">
+    <#if tc.architectureOutputs?size gt 1>
+        outputs.append(${input})
+    <#else>
         return ${input}
+    </#if>
 </#if>

src/test/java/de/monticore/lang/monticar/cnnarch/gluongenerator/GenerationTest.java

@@ -30,6 +30,8 @@ import de.se_rwth.commons.logging.Finding;
 import de.se_rwth.commons.logging.Log;