add protocols for other users and add a seed fixing util

README.md

@@ -1,4 +1,4 @@
-# Util Collections
+# Utils Collection
 
 ## Motivation
 When we implement codes, we often search for util functions that are already implemented. 
@@ -8,6 +8,23 @@ Here, we are going to share util functions that are often used by others.
 - You should provide a code in folder and short description about the code. 
 - Never push on master. We have prohibited direct push on main branch. Use pull request to avoid confusion. 
 - After adding an util function, add an item in the list below. 
+- Please refer to other formats. 
+- It is not necessary to be a python code. Feel free to add util functions. 
+- We strongly recommend to add the description as the following format: 
+
+Example 
+```
+Author: YoonwooJeong
+Description: Example for Addition
+Input (Optional): 
+    torch.Tensor a : the first number to add
+    torch.Tensor b : the second number to add
+Output (Optional): 
+    torch.Tensor out: the addition of "a" and "b". 
+Requirements (Optional): 
+    numpy >= 0.0.1, torch >= 0.0.1
+```
 
 ## Util Functions
-- Fixing the seed
+- Fixing the seed (torch_function/fix_seed.py)
+    - A function to completely fix the seeds of torch, numpy, and random.

description_format.txt

@@ -0,0 +1,7 @@
+"""
+Author: 
+Description: 
+Input (Optional): 
+Output (Optional): 
+Requirements (Optional)
+"""

torch_function/fix_seed.py

@@ -0,0 +1,23 @@
+import numpy as np
+import torch
+import random
+"""
+Author: Boseung Jeong and Yoonwoo Jeong
+Description: Fixing the seed. When deterministic is true, the seed is 
+    completely fixed, but the speed of programs might be slower.
+Input (Optional): 
+    int seed : seed to fix
+    bool deterministic: fix the torch.backends.cudnn.deterministic
+Output (Optional): None
+Requirements (Optional): 
+    numpy, torch, random
+"""
+def fix_seed(seed, deterministic=True):
+    np.random.seed(seed)
+    torch.manual_seed(seed)
+    torch.cuda.manual_seed(seed)
+    torch.cuda.manual_seed_all(seed) # if use multi-GPU
+    torch.backends.cudnn.deterministic = deterministic
+    torch.backends.cudnn.benchmark = False
+    np.random.seed(seed)
+    random.seed(seed)