diff --git a/mhcflurry/class1_neural_network.py b/mhcflurry/class1_neural_network.py
index 92cf8b27ebe74e3924cfe891f139a35cc0a84c51..c1f8e106820a47dcddd76ecae350191f64716121 100644
--- a/mhcflurry/class1_neural_network.py
+++ b/mhcflurry/class1_neural_network.py
@@ -718,6 +718,7 @@ class Class1NeuralNetwork(object):
                     y_values,
                 ]),
             }
+            adjusted_inequalities_with_random_negatives = None
         if sample_weights is not None:
             sample_weights_with_random_negatives = numpy.concatenate([
                 numpy.ones(int(num_random_negative.sum())),
diff --git a/mhcflurry/custom_loss.py b/mhcflurry/custom_loss.py
index 641fd09969700670b22128059ef078151cf7d30e..eabdd130a99dcbd828b58a9f2c327832bbf18eeb 100644
--- a/mhcflurry/custom_loss.py
+++ b/mhcflurry/custom_loss.py
@@ -82,7 +82,7 @@ class MSEWithInequalities(Loss):
     def encode_y(y, inequalities=None):
         y = array(y, dtype="float32")
         if isnan(y).any():
-            raise ValueError("y contains NaN")
+            raise ValueError("y contains NaN: %s" % str(y))
         if (y > 1.0).any():
             raise ValueError("y contains values > 1.0")
         if (y < 0.0).any():
@@ -141,7 +141,7 @@ class MSEWithInequalitiesAndMultipleOutputs(Loss):
     def encode_y(y, inequalities=None, output_indices=None):
         y = array(y, dtype="float32")
         if isnan(y).any():
-            raise ValueError("y contains NaN")
+            raise ValueError("y contains NaN: %s" % str(y))
         if (y > 1.0).any():
             raise ValueError("y contains values > 1.0")
         if (y < 0.0).any():