diff --git a/train.py b/train.py
index 5c160e8..176b30c 100644
--- a/train.py
+++ b/train.py
@@ -86,11 +86,19 @@ def get_inductive_links(df, train_edge_end, val_edge_end):
 
 if args.use_inductive:
     test_df = df[val_edge_end:]
-    inductive_nodes = set(test_df.src.values).union(test_df.src.values)
+    train_df = df[:train_edge_end]
+    train_nodes = set(train_df.src.values).union(train_df.dst.values)
+    all_nodes = set(df.src.values).union(df.dst.values)
+    inductive_nodes = set(test_df.src.values).union(test_df.dst.values)
     print("inductive nodes", len(inductive_nodes))
-    neg_link_sampler = NegLinkInductiveSampler(inductive_nodes)
+    neg_link_sampler_train = NegLinkInductiveSampler(train_nodes, seed=args.exp_seed)
+    neg_link_sampler_val = NegLinkInductiveSampler(all_nodes, seed=args.exp_seed)
+    neg_link_sampler_test = NegLinkInductiveSampler(inductive_nodes, seed=args.exp_seed)
 else:
     neg_link_sampler = NegLinkSampler(g['indptr'].shape[0] - 1)
+    neg_link_sampler_train = neg_link_sampler
+    neg_link_sampler_val = neg_link_sampler
+    neg_link_sampler_test = neg_link_sampler
 
 def eval(mode='val'):
     neg_samples = 1
@@ -98,11 +106,14 @@ def eval(mode='val'):
     aps = list()
     aucs_mrrs = list()
     if mode == 'val':
+        neg_link_sampler = neg_link_sampler_val
         eval_df = df[train_edge_end:val_edge_end]
     elif mode == 'test':
+        neg_link_sampler = neg_link_sampler_test
         eval_df = df[val_edge_end:]
         neg_samples = args.eval_neg_samples
     elif mode == 'train':
+        neg_link_sampler = neg_link_sampler_train
         eval_df = df[:train_edge_end]
     with torch.no_grad():
         total_loss = 0
@@ -189,7 +200,7 @@ def eval(mode='val'):
         model.memory_updater.last_updated_nid = None
     for _, rows in df[:train_edge_end].groupby(group_indexes[random.randint(0, len(group_indexes) - 1)]):
         t_tot_s = time.time()
-        root_nodes = np.concatenate([rows.src.values, rows.dst.values, neg_link_sampler.sample(len(rows))]).astype(np.int32)
+        root_nodes = np.concatenate([rows.src.values, rows.dst.values, neg_link_sampler_train.sample(len(rows))]).astype(np.int32)
         ts = np.concatenate([rows.time.values, rows.time.values, rows.time.values]).astype(np.float32)
         if sampler is not None:
             if 'no_neg' in sample_param and sample_param['no_neg']: