diff --git a/keras_wrapper/extra/evaluation.py b/keras_wrapper/extra/evaluation.py
index 5112788..f845236 100644
--- a/keras_wrapper/extra/evaluation.py
+++ b/keras_wrapper/extra/evaluation.py
@@ -249,7 +249,7 @@ def multiclass_metrics(pred_list, verbose, extra_vars, split):
     accuracy = sklearn_metrics.accuracy_score(y_gt, y_pred)
     acc_top_n = {}
     for topn in top_n_accuracies:
-        acc_top_n[topn] = __top_k_accuracy(y_gt, y_pred, topn)
+        acc_top_n[topn] = __top_k_accuracy(y_gt, pred_list, topn)
     # accuracy_balanced = sklearn_metrics.accuracy_score(y_gt, y_pred, sample_weight=sample_weights, )
 
     # The following two lines should both provide the same measure (balanced accuracy)
@@ -287,7 +287,7 @@ def __top_k_accuracy(truths, preds, k):
     :param k:
     :return:
     """
-    best_k = np.argsort(preds, axis=1)[:, -k:]
+    best_k = np.argsort(preds, axis=1)[:, -k:][:, ::-1]
     ts = np.argmax(truths, axis=1)
     successes = 0
     for i in range(ts.shape[0]):