diff --git a/README.md b/README.md
index 044d7e0..277514e 100644
--- a/README.md
+++ b/README.md
@@ -9,7 +9,7 @@ This repository provides a reference implementation for learning NeuralSort-base
 
 ## Requirements
 
-The codebase is implemented in Python 3.7. To install the necessary requirements, run the following commands:
+The codebase is implemented in Python 3.6. To install the necessary requirements, run the following commands:
 
 ```
 pip install -r requirements.txt
diff --git a/tf/run_median.py b/tf/run_median.py
index 05f4808..e98a2b0 100644
--- a/tf/run_median.py
+++ b/tf/run_median.py
@@ -231,7 +231,7 @@ def prnt(*args):
 TRAIN_PER_EPOCH = mnist_input.TRAIN_SET_SIZE // (l * M)
 VAL_PER_EPOCH = mnist_input.VAL_SET_SIZE // (l * M)
 TEST_PER_EPOCH = mnist_input.TEST_SET_SIZE // (l * M)
-best_val = float('inf')
+best_correct_val = float('inf')
 tiebreaker_val = -1
 
 
@@ -272,8 +272,8 @@ def test(epoch, val=False):
     if val:
         prnt("Validation set: correctly identified %f, mean squared error %f" %
              (c_i, l_v))
-        if l_v < best_val:
-            best_val = l_v
+        if l_v < best_correct_val:
+            best_correct_val = l_v
             prnt('Saving...')
             save_model(epoch)
     else:
diff --git a/tf/run_sort.py b/tf/run_sort.py
index 0b4790c..04ad7da 100644
--- a/tf/run_sort.py
+++ b/tf/run_sort.py
@@ -73,7 +73,7 @@
     losses = tf.reduce_mean(losses, axis=-1)
     loss = tf.reduce_mean(losses)
 
-if method == 'sinkhorn':
+elif method == 'sinkhorn':
     representations = multi_mnist_cnn.deepnn(l, X, n)
     pre_sinkhorn = tf.reshape(representations, [M, n, n])
     P_hat = sinkhorn_operator(pre_sinkhorn, temp=temperature)
@@ -84,7 +84,7 @@
     losses = tf.reduce_mean(losses, axis=-1)
     loss = tf.reduce_mean(losses)
 
-if method == 'gumbel_sinkhorn':
+elif method == 'gumbel_sinkhorn':
     representations = multi_mnist_cnn.deepnn(l, X, n)
     pre_sinkhorn = tf.reshape(representations, [M, n, n])
     P_hat = sinkhorn_operator(pre_sinkhorn, temp=temperature)
@@ -102,7 +102,7 @@
     losses = tf.reshape(losses, [-1])
     loss = tf.reduce_mean(losses)
 
-if method == 'deterministic_neuralsort':
+elif method == 'deterministic_neuralsort':
     scores = multi_mnist_cnn.deepnn(l, X, 1)
     scores = tf.reshape(scores, [M, n, 1])
     P_hat = util.neuralsort(scores, temperature)
@@ -112,7 +112,7 @@
     losses = tf.reduce_mean(losses, axis=-1)
     loss = tf.reduce_mean(losses)
 
-if method == 'stochastic_neuralsort':
+elif method == 'stochastic_neuralsort':
     scores = multi_mnist_cnn.deepnn(l, X, 1)
     scores = tf.reshape(scores, [M, n, 1])
     P_hat = util.neuralsort(scores, temperature)