First Testing Completed, mark version, further tests pending.

code.py

@@ -21,6 +21,7 @@ def custom_ops(n):
     required_lower_triangle = tf.reverse(transposed_lower_triangle, [0])
     reverse_of_mat_minus_required = tf.subtract(reverse_of_mat, lower_triangular_without_diagonal)
     mat_minus_required = tf.reverse(reverse_of_mat_minus_required, [1])
+
     step_1 = tf.add(required_lower_triangle, mat_minus_required)
 
     # Step 2: Take the maximum value along the columns of A to get a vector m
@@ -41,34 +42,37 @@ def custom_ops(n):
 
     with_nan = lower_triangular_without_diagonal_inf_m = tf.subtract(inf_m, right_lower_triangular_inf_m)
     replaced_zero = tf.where(tf.is_nan(with_nan), tf.zeros([n, n]), with_nan)
+
     to_softmax = tf.subtract(reverse_upper_triangular_reverse_m, replaced_zero)
 
-    step_3 = tf.nn.softmax(to_softmax)
+    step_3 = B = tf.nn.softmax(to_softmax)
 
     # Step 4: Sum along the rows of B to obtain vector v1
 
-    step_4 = tf.reduce_sum(step_3, 1)
+    step_4 = v1 = tf.reduce_sum(B, 1)
 
     # Step 5: Sum along the columns of B to get another vector v2
 
-    step_5 = tf.reduce_sum(step_3, 0)
+    step_5 = v2 = tf.reduce_sum(B, 0)
 
     # Step 6: Concatenate the two vectors and take a softmax of this vector
 
-    concat = tf.concat([step_4, step_5], 0)
+    concat = tf.concat([v1, v2], 0)
+
     step_6 = v = tf.nn.softmax(concat)
 
     # Step 7: Get the index number in vector v with maximum value
 
-    step_7 = tf.to_float(tf.argmax(step_6))
+    step_7 = index_number = tf.to_float(tf.argmax(v))
 
     # Step 8: If the index number is greater than n/3 store:
     #         finalVal = ||v1 − v2||2
     #         Otherwise, store:
     #         finalVal = ||v1 + v2||2
 
     n_by_3 = tf.constant(n/3)
-    finalVal = tf.cond(step_7 > n_by_3, lambda: tf.norm(step_4 - step_5), lambda: tf.norm(step_4 + step_5))
+
+    finalVal = tf.cond(index_number > n_by_3, lambda: tf.norm(v1 - v2), lambda: tf.norm(v1 + v2))
 
     # Step 9: Return the variable finalVal