diff --git a/algo/curve_anomaly/cont_det/cont_device.py b/algo/curve_anomaly/cont_det/cont_device.py
index ab8a687..8ac79a8 100644
--- a/algo/curve_anomaly/cont_det/cont_device.py
+++ b/algo/curve_anomaly/cont_det/cont_device.py
@@ -118,7 +118,11 @@ def get_reconstruction_errors(model_input_data_array, model, use_cuda):
         model_input = torch.Tensor(data_series)
         if use_cuda:
             model_input = model_input.cuda()
-        model_output = model(model_input)
+        try:
+            model_output = model(model_input)
+        except RuntimeError:
+            print("Probably not enough data collected yet!")
+            return None, None
         errors.append(abs(model_output.detach().cpu().numpy()-data_series).sum()/205)
     model.train()
     return errors, model_output.detach().cpu().numpy()
@@ -134,6 +138,9 @@ def test(data_list, model, use_cuda, anomalies, training_max, reconstruction_err
     model_input_data_array = np.array(data_series_smooth[-model_input_window_length:]).reshape(1,-1)
     new_reconstruction_error = get_reconstruction_errors(model_input_data_array, model, use_cuda)[0][0]
     reconstructed_curve = get_reconstruction_errors(model_input_data_array, model, use_cuda)[1].flatten()
+    if new_reconstruction_error == None: # This happens if not enough data is collected yet!
+        model.train()
+        return None, anomalies, reconstruction_errors
     if reconstruction_errors == None:
         reconstruction_errors = [new_reconstruction_error]
     else: