learn_to_drive: add up threshold in reLU to avoid NaN in neuron weight calc

deepQlearn_0/src/deepQlearning/learn_to_drive.c

@@ -2,14 +2,16 @@
 
 char *action_name[8] = {"LEFT", "CENTER", "RIGHT"};
 
+#define THRESHOLD_UP 10
+
 float reLU(float x){
-//  if(x>10) return 10;
+  if(x>THRESHOLD_UP) return THRESHOLD_UP;
   if(x>0) return x;
   return 0;
 }
 
 float d_reLU(float x){
-//  if (x>10) return 0;
+  if (x>THRESHOLD_UP) return 0;
   if (x>0) return 1;
   return 0;
 }