Add support for “probs” to mlp. Fixes =O_032=

acletournel · Feb 16, 2019 · 7ed3974 · 7ed3974
1 parent 4e5e931
commit 7ed3974
Showing 1 changed file with 23 additions and 10 deletions.
diff --git a/sources/regression/ml_mlp.cpp b/sources/regression/ml_mlp.cpp
@@ -86,7 +86,8 @@ namespace ml
             FLEXT_CADDATTR_SET(c, "use_validation_set", set_use_validation_set);
             FLEXT_CADDATTR_SET(c, "validation_set_size", set_validation_set_size);
             FLEXT_CADDATTR_SET(c, "randomize_training_order", set_randomise_training_order);
-
+            FLEXT_CADDATTR_SET(c, "probs", set_probs);
+
             FLEXT_CADDATTR_GET(c, "mode", get_mode);
             FLEXT_CADDATTR_GET(c, "num_outputs", get_num_outputs);
             FLEXT_CADDATTR_GET(c, "num_hidden", get_num_hidden);
@@ -104,7 +105,8 @@ namespace ml
             FLEXT_CADDATTR_GET(c, "use_validation_set", get_use_validation_set);
             FLEXT_CADDATTR_GET(c, "validation_set_size", get_validation_set_size);
             FLEXT_CADDATTR_GET(c, "randomize_training_order", get_randomise_training_order);
-
+            FLEXT_CADDATTR_GET(c, "probs", get_probs);
+
             DefineHelp(c, object_name.c_str());
         }
 
@@ -132,7 +134,8 @@ namespace ml
         void set_use_validation_set(bool use_validation_set);
         void set_validation_set_size(int validation_set_size);
         void set_randomise_training_order(bool randomise_training_order);
-
+        void set_probs(bool probs);
+
         // Flext attribute getters
         void get_mode(int &mode) const;
         void get_num_outputs(int &num_outputs) const;
@@ -152,7 +155,8 @@ namespace ml
         void get_use_validation_set(bool &use_validation_set) const;
         void get_validation_set_size(int &validation_set_size) const;
         void get_randomise_training_order(bool &randomise_training_order) const;
-
+        void get_probs(bool &probs) const;
+
         // Implement pure virtual methods
         GRT::MLBase &get_MLBase_instance();
         const GRT::MLBase &get_MLBase_instance() const;
@@ -184,6 +188,7 @@ namespace ml
         FLEXT_CALLVAR_B(get_use_validation_set, set_use_validation_set);
         FLEXT_CALLVAR_I(get_validation_set_size, set_validation_set_size);
         FLEXT_CALLVAR_B(get_randomise_training_order, set_randomise_training_order);
+        FLEXT_CALLVAR_B(get_probs, set_probs);
 
         // Virtual method override
         virtual const std::string get_object_name(void) const { return object_name; };
@@ -194,6 +199,7 @@ namespace ml
         GRT::Neuron::Type hidden_activation_function;
         GRT::Neuron::Type output_activation_function;
 
+        bool probs;
     };
 
     // Flext attribute setters
@@ -422,6 +428,11 @@ namespace ml
         }
     }
 
+    void mlp::set_probs(bool probs)
+    {
+        this->probs = probs;
+    }
+
     // Flext attribute getters
     void mlp::get_mode(int &mode) const
     {
@@ -520,6 +531,9 @@ namespace ml
     void mlp::get_randomise_training_order(bool &randomise_training_order) const
     {
         flext::error("function not implemented");
+    void mlp::get_probs(bool &probs) const
+    {
+        probs = this->probs;
     }
 
     // Methods
@@ -620,7 +634,6 @@ namespace ml
             return;
         }
 
-        // TODO: add probs to attributes
         if (grt_mlp.getClassificationModeActive())
         {
             GRT::VectorDouble likelihoods = grt_mlp.getClassLikelihoods();
@@ -631,9 +644,9 @@ namespace ml
             {
                 flext::error("labels / likelihoods size mismatch");
             }
-            else
+            else if (probs)
             {
-                AtomList probs;
+                AtomList probs_list;
 
                 for (unsigned count = 0; count < labels.size(); ++count)
                 {
@@ -643,10 +656,10 @@ namespace ml
                     SetFloat(likelihood_a, static_cast<float>(likelihoods[count]));
                     SetInt(label_a, labels[count]);
 
-                    probs.Append(label_a);
-                    probs.Append(likelihood_a);
+                    probs_list.Append(label_a);
+                    probs_list.Append(likelihood_a);
                 }
-                ToOutAnything(1, get_s_probs(), probs);
+                ToOutAnything(1, get_s_probs(), probs_list);
             }
 
             ToOutInt(0, classification);