如何消除Kaggle Kernel中全局/操作级种子对后续单元格的影响？

问题背景 #

我在编写涉及全局/操作级种子设置的Kaggle Kernel时，将每个测试案例放在独立单元格中，但发现前序单元格的种子设置会影响后续单元格的输出。

比如下面这段代码，作为独立Python文件运行时，每次都会生成完全相同的序列：

"""
Case:
    Setup: When only operation level seed is set -
           - Since operation level seed is set, the sequence will start from same number after restarts, 
            and sequence will be same

"""

import tensorflow as tf

print(tf.random.uniform([1], seed=1))  # generates 'A1'
print(tf.random.uniform([1], seed=1))  # generates 'A2'
print(tf.random.uniform([1], seed=1))  # generates 'A3'
print(tf.random.uniform([1], seed=1))  # generates 'A4'

但把这段代码放在Kaggle Kernel的独立单元格中重复运行时，得到的序列却不一致，推测是前序单元格的全局/操作级种子残留影响导致的。

已尝试的解决方案 #

我试过两种重启内核的方式来消除前序单元格的影响，但问题依然存在：

第一种方法 #

import IPython.display as display
import time

# Restart the cell
display.Javascript('Jupyter.notebook.kernel.restart()')

# Wait for the kernel to restart
time.sleep(20)

"""
Case:
    Setup: When only operation level seed is set -
           - Since operation level seed is set, the sequence will start from same number after restarts, 
            and sequence will be same

"""

import tensorflow as tf

print(tf.random.uniform([1], seed=1))  # generates 'A1'
print(tf.random.uniform([1], seed=1))  # generates 'A2'
print(tf.random.uniform([1], seed=1))  # generates 'A3'
print(tf.random.uniform([1], seed=1))  # generates 'A4'

第二种方法 #

from IPython.core.display import HTML
HTML("<script>Jupyter.notebook.kernel.restart()</script>")

"""
Case:
    Setup: When only operation level seed is set -
           - Since operation level seed is set, the sequence will start from same number after restarts, 
            and sequence will be same

"""

import tensorflow as tf

print(tf.random.uniform([1], seed=1))  # generates 'A1'
print(tf.random.uniform([1], seed=1))  # generates 'A2'
print(tf.random.uniform([1], seed=1))  # generates 'A3'
print(tf.random.uniform([1], seed=1))  # generates 'A4'

提问 #

请问还有哪些方法可以尝试？或是我的操作中存在什么错误？

内容的提问来源于stack exchange，提问作者Deepak Tatyaji Ahire