diff --git a/config.yaml b/config.yaml
index 2a67f10..03b6560 100644
--- a/config.yaml
+++ b/config.yaml
@@ -1,7 +1,7 @@
 # config for shadow
 CFG:
   DEBUG: false
-  num_accessible_probs: 10 # match with accessible classes logits/probability classes number from the target model. usually top 5 for APIs
+  topk_num_accessible_probs: 10 # topk match with accessible classes logits/probability classes number from the target model. usually top 5 for APIs
   num_epochs: 25 # number of shadow model train epochs
   val_acc_goal: -1 # shadow model's goal accuracy working as early stop. -1 for not using early stop
   num_shadow_models: 100
diff --git a/shadow/make_data.py b/shadow/make_data.py
index 5b08e48..709cbd6 100644
--- a/shadow/make_data.py
+++ b/shadow/make_data.py
@@ -39,7 +39,7 @@ def make_member_nonmember(finetuned_model, trainloader, valloader, criterion, de
             prob = F.softmax(output, dim=1)  # softmax logits
 
             # get top inferred classes probability and append to member_dset
-            top_p, top_class = prob.topk(CFG.num_accessible_probs, dim=1)
+            top_p, top_class = prob.topk(CFG.topk_num_accessible_probs, dim=1)
             top_p = top_p.cpu().detach().numpy()  # detach from cuda
             # loss = criterion(output, labels)
             member_dset.append(top_p)
@@ -53,7 +53,7 @@ def make_member_nonmember(finetuned_model, trainloader, valloader, criterion, de
             prob = F.softmax(output, dim=1)  # softmax logits
 
             # get top inferred classes probability and append to member_dset
-            top_p, top_class = prob.topk(CFG.num_accessible_probs, dim=1)
+            top_p, top_class = prob.topk(CFG.topk_num_accessible_probs, dim=1)
             top_p = top_p.cpu().detach().numpy()  # detach from cuda
 
             # append to non_member_dset
diff --git a/train_shadow.py b/train_shadow.py
index 8c1061a..26e31ed 100644
--- a/train_shadow.py
+++ b/train_shadow.py
@@ -51,7 +51,7 @@
 
 # define dataset for attack model that shadow models will generate
 print("mapped classes to ids:", trainset.class_to_idx)
-columns_attack_sdet = [f"top_{index}_prob" for index in range(CFG.num_accessible_probs)]
+columns_attack_sdet = [f"top_{index}_prob" for index in range(CFG.topk_num_accessible_probs)]
 df_attack_dset = pd.DataFrame({}, columns=columns_attack_sdet + ["is_member"])
 
 # random subset for shadow model train & validation from the CIFAR trainset