CS 20 : TensorFlow for Deep Learning Research¶

Randomization¶

Setup¶

In [1]:

from __future__ import absolute_import, division, print_function
import tensorflow as tf

print(tf.__version__)

1.12.0

Example 1 : Session keeps track of the random state¶

In [2]:

c = tf.random_normal(shape = [], seed = 2)

with tf.Session() as sess:
    print(sess.run(c))
    print(sess.run(c))

-0.85811085
-0.20793143

Example 2 : Each new session will start the random state all over again¶

In [3]:

tf.reset_default_graph()

c = tf.random_normal(shape = [], seed = 2)

with tf.Session() as sess:
    print(sess.run(c))
    
with tf.Session() as sess:
    print(sess.run(c))
    print(sess.run(c))

-0.85811085
-0.85811085
-0.20793143

Example 3 : With operation level random seed, each op keeps its own seed¶

In [4]:

tf.reset_default_graph()

c = tf.random_normal(shape = [], seed = 2)
d = tf.random_normal(shape = [], seed = 2)

with tf.Session() as sess:
    print(sess.run(c))
    print(sess.run(d))

-0.85811085
-0.85811085

Example 4 : Graph level random seed¶

In [5]:

tf.reset_default_graph()
tf.set_random_seed(seed = 2)

c = tf.random_normal(shape = [])
d = tf.random_normal(shape = [])

with tf.Session() as sess:
    print(sess.run(c))
    print(sess.run(d))

-1.4236197
0.8052349

In [6]:

tf.reset_default_graph()
tf.set_random_seed(seed = 2)

c = tf.random_normal(shape = [])
d = tf.random_normal(shape = [])

with tf.Session() as sess:
    print(sess.run(c))
    print(sess.run(d))

-1.4236197
0.8052349

CS 20 : TensorFlow for Deep Learning Research¶

Lecture 05 : Variable sharing and managing experiments¶

Randomization¶

Setup¶

Example 1 : Session keeps track of the random state¶

Example 2 : Each new session will start the random state all over again¶

Example 3 : With operation level random seed, each op keeps its own seed¶

Example 4 : Graph level random seed¶