Added a way to ensure neural network shape is not changing during decoding

maca_trans_parser/src/maca_tm_train.c

@@ -211,7 +211,6 @@ void train_nn(context * ctx)
     classifier_print_desc_file(classif->filename, classif);
 
     classifier_delete_mlp(classif);
-    //classifier_init_mlp(classif, new Mlp(classif->mlp_model_filename, classif->mlp_struct_filename, ctx->maca_data_path));
   }
 }
 

mlp/include/mlp.h

@@ -69,7 +69,6 @@ class Mlp{
 
   Mlp(std::vector<Layer> layers, unsigned int batch_size = 0);
   Mlp(char * model_filename, char * struct_filename, char * absolute_path);
-  unsigned int predict(int * features, int nb_features);
   unsigned int predict(float * features, int nb_features);
   void train(int nb_iter_max, Fann_file & fann_train, Fann_file & fann_dev,
              std::function<void(std::vector< std::vector<float> >&,

mlp/src/mlp.cpp

@@ -162,21 +162,19 @@ unsigned int Mlp::predict(dynet::Expression x){
   return argmax;
 }
 
-unsigned int Mlp::predict(int * features, int nb_features){
+unsigned int Mlp::predict(float * features, int nb_features){
   computation_graph.clear();
 
   unsigned int nb_inputs = layers[0].input_dim;
   std::vector<float> features_vector(features, features + nb_features);
-  dynet::Expression x = input(computation_graph, {nb_inputs}, features_vector);
 
-  return predict(x);
+  if(features_vector.size() != nb_inputs)
+  {
+    fprintf(stderr, "ERROR (%s) : example size=%lu nb_inputs=%u mismatch\n",
+            __func__, features_vector.size(), nb_inputs);
+    exit(1);
   }
 
-unsigned int Mlp::predict(float * features, int nb_features){
-  computation_graph.clear();
-
-  unsigned int nb_inputs = layers[0].input_dim;
-  std::vector<float> features_vector(features, features + nb_features);
   dynet::Expression x = input(computation_graph, {nb_inputs}, features_vector);
 
   return predict(x);