diff --git a/python/utils/tasks.py b/python/utils/tasks.py
index 58e8d1a..465512f 100644
--- a/python/utils/tasks.py
+++ b/python/utils/tasks.py
@@ -43,8 +43,8 @@ def dist_matrix(D1,D2,distance):
     if distance=='L2':
         if TORCH_GPU_IS_AVAILABLE:
             with torch.no_grad():
-                D = torch.cdist(torch.from_numpy(D1).cuda(),
-                                torch.from_numpy(D2).cuda(),
+                D = torch.cdist(torch.from_numpy(D1.astype(float)).cuda(),
+                                torch.from_numpy(D2.astype(float)).cuda(),
                                 p = 2).detach().cpu().numpy()
         else:
             D = spatial.distance.cdist(D1, D2,'euclidean')
@@ -96,9 +96,9 @@ def eval_verification(descr,split):
     print('>> Evaluating %s task' % green('verification'))
 
     start = time.time()
-    pos = pd.read_csv(os.path.join(tskdir, 'verif_pos_split-'+split['name']+'.csv')).as_matrix()
-    neg_intra = pd.read_csv(os.path.join(tskdir, 'verif_neg_intra_split-'+split['name']+'.csv')).as_matrix()
-    neg_inter = pd.read_csv(os.path.join(tskdir, 'verif_neg_inter_split-'+split['name']+'.csv')).as_matrix()
+    pos = pd.read_csv(os.path.join(tskdir, 'verif_pos_split-'+split['name']+'.csv')).to_numpy()
+    neg_intra = pd.read_csv(os.path.join(tskdir, 'verif_neg_intra_split-'+split['name']+'.csv')).to_numpy()
+    neg_inter = pd.read_csv(os.path.join(tskdir, 'verif_neg_inter_split-'+split['name']+'.csv')).to_numpy()
 
     d_pos = get_verif_dists(descr,pos,1)
     d_neg_intra = get_verif_dists(descr,neg_intra,2)
@@ -236,8 +236,8 @@ def eval_retrieval(descr,split): #WIP
     print('>> Evaluating %s task' % green('retrieval'))
     start = time.time()
 
-    q = pd.read_csv(os.path.join(tskdir, 'retr_queries_split-'+split['name']+'.csv')).as_matrix()
-    d = pd.read_csv(os.path.join(tskdir, 'retr_distractors_split-'+split['name']+'.csv')).as_matrix()
+    q = pd.read_csv(os.path.join(tskdir, 'retr_queries_split-'+split['name']+'.csv')).to_numpy()
+    d = pd.read_csv(os.path.join(tskdir, 'retr_distractors_split-'+split['name']+'.csv')).to_numpy()
 
     # q_std = np.std(q, axis=0)
     # d_std = np.std(d, axis=0)