Minor fix in batchnorm_layer (side effect in python samples)

src/batchnorm_layer.c

@@ -182,9 +182,9 @@ void forward_batchnorm_layer_gpu(layer l, network_state state)
     if (l.type == BATCHNORM) simple_copy_ongpu(l.outputs*l.batch, state.input, l.output_gpu);
         //copy_ongpu(l.outputs*l.batch, state.input, 1, l.output_gpu, 1);
 
-    simple_copy_ongpu(l.outputs*l.batch, l.output_gpu, l.x_gpu);
-    //copy_ongpu(l.outputs*l.batch, l.output_gpu, 1, l.x_gpu, 1);
     if (state.train) {
+        simple_copy_ongpu(l.outputs*l.batch, l.output_gpu, l.x_gpu);
+        //copy_ongpu(l.outputs*l.batch, l.output_gpu, 1, l.x_gpu, 1);
 #ifdef CUDNN
         float one = 1;
         float zero = 0;

src/parser.c

@@ -978,6 +978,7 @@ network parse_network_cfg_custom(char *filename, int batch, int time_steps)
     params.batch = net.batch;
     params.time_steps = net.time_steps;
     params.net = net;
+    printf("batch = %d, time_steps = %d, train = %d \n", net.batch, net.time_steps, params.train);
 
     float bflops = 0;
     size_t workspace_size = 0;