TensorFlow はじめの一歩(2) - みらいテックラボ

今回は, 定数/数列/乱数テンソル[1]に関するPython APIについて, 少しまとめておく.
注) TensorFlow : Python API 0.6.0ベース

API	機能
tf.zeros(shape, dtype=tf.float32, name=None)	すべての要素が0のTensorを生成
tf.zeros_like(tensor, dtype=None, name=None)	tensorと同じtype, shapeで, すべての要素が0のTensorを生成
tf.ones(shape, dtype=tf.float32, name=None)	すべての要素が1のTensorを生成
tf.ones_like(tensor, dtype=None, name=None)	tensorと同じtype, shapeで, すべての要素が1のTensorを生成
tf.fill(dims, value, name=None)	スカラー値でfillしたTensorを生成
tf.constant(value, dtype=None, shape=None, name=Const)	定数のTensorを生成

サンプルコード：

import tensorflow as tf

sess = tf.InteractiveSession()

################
# tf.zeros
################
tf_zeros = tf.zeros([3, 4], "int32")
print "tf.zeros"
print sess.run(tf_zeros)

# output:

# tf.zeros
# [[0 0 0 0]
#  [0 0 0 0]
#  [0 0 0 0]]

################
# tf.zeros_like
################
a = tf.constant([[1, 2, 3], [4, 5, 6]])
tf_zeros_like = tf.zeros_like(a)
print "a"
print sess.run(a)
print "tf.zeros_like"
print sess.run(tf_zeros_like)

# output:

# a
# [[1 2 3]
#  [4 5 6]]
# tf.zeros_like
# [[0 0 0]
#  [0 0 0]]

################
# tf.ones
################
tf_ones = tf.ones([2, 3], "int32")
print "tf.ones"
print sess.run(tf_ones)

# output:

# tf.ones
# [[1 1 1]
#  [1 1 1]]

################
# tf.ones_like
################
a = tf.constant([[1, 2, 3], [4, 5, 6]])
tf_ones_like = tf.ones_like(a)
print "a"
print sess.run(a)
print "tf.ones_like"
print sess.run(tf_ones_like)

# output:

# a
# [[1 2 3]
#  [4 5 6]]
# tf.ones_like
# [[1 1 1]
#  [1 1 1]]

################
# tf.fill
################
tf_fill = tf.fill((2, 3), 9)
print "tf.fill"
print sess.run(tf_fill)

# output:

# tf.fill
# [[9 9 9]
#  [9 9 9]]

################
# tf.constant
################
# 1-D Tensor
tf_const1 = tf.constant([1, 2, 3, 4, 5, 6, 7])
# 2-D Tensor
tf_const2 = tf.constant(-1.0, shape=[2, 3])
print "tf.constant"
print "1-D"
print sess.run(tf_const1)
print "2-D"
print sess.run(tf_const2)

# output:

# tf.constant
# 1-D
# [1 2 3 4 5 6 7]
# 2-D
# [[-1. -1. -1.]
#  [-1. -1. -1.]]

sess.close()

2. 数列

API	機能
tf.linspace(start, stop, num, name=None)	ある範囲を要素数になるように区切った値を生成
tf.range(start, limit=None, delta=1, name=range)	一定間隔の整数列を生成

サンプルコード:

import tensorflow as tf

sess = tf.InteractiveSession()

################
# tf.linspace
################
tf_linspace = tf.linspace(10.0, 12.0, 3, name="linspace")
print "tf.linspace"
print sess.run(tf_linspace)

# output:

# tf.linspace
# [ 10.  11.  12.]

################
# tf.range
################
# 'start' is 3
# 'limit' is 18
# 'delta' is 3
tf_range1 = tf.range(3, 18, 3)
# 'limit' is 5
tf_range2 = tf.range(5)
print "tf.range"
print "start = 3, limit = 18, delta = 3"
print sess.run(tf_range1)
print "limit = 5"
print sess.run(tf_range2)

# output:

# tf.range
# start = 3, limit = 18, delta = 3
# [ 3  6  9 12 15]
# limit = 5
# [0 1 2 3 4]

sess.close()

3. 乱数テンソル

API	機能
tf.random_normal(shape, mean=0.0, stddev=1.0, dtype=tf.float32, seed=None, name=None)	正規分布から乱数を出力
tf.truncated_normal(shape, mean=0.0, stddev=1.0, dtype=tf.float32, seed=None, name=None)	切断正規分布から乱数を出力
tf.random_uniform(shape, minval=0, maxval=None, dtype=tf.float32, seed=None, name=None)	一様分布から乱数を出力
tf.random_shuffle(value, seed=None, name=None)	Tensorをランダムにシャッフル
tf.set_random_seed(seed)	乱数のシード値を設定

サンプルコード:

import tensorflow as tf

sess = tf.InteractiveSession()

######################
# tf.random_normal
######################
tf_normal = tf.random_normal([2, 3], mean=-1, stddev=4)
print "tf.random_normal"
print sess.run(tf_normal)

# output(example):

# tf.random_normal
# [[-4.77949524  1.09623837 -1.79200745]
#  [ 0.96336484 -5.47785091 -1.14655089]

######################
# tf.truncated_normal
######################
tf_truncated = tf.truncated_normal([2, 3])
print "tf.truncated_normal"
print sess.run(tf_truncated)

# output(example):

# tf.truncated_normal
# [[ 0.73809606  1.65349591  0.15235977]
#  [ 0.42829624 -0.83115262 -0.44399944]]

######################
# tf.random_uniform
######################
tf_uniform = tf.random_uniform([2, 3], minval=0, maxval=5)
print "tf.random_uniform"
print sess.run(tf_uniform)

# output(example):

# tf.random_uniform
# [[ 1.65245831  4.85745525  1.04827166]
#  [ 2.29264975  0.33302426  1.61023617]]

######################
# tf.random_shuffle
######################
c = tf.constant([[1, 2], [3, 4], [5, 6]])
tf_shuff = tf.random_shuffle(c)
print "c"
print sess.run(c)
print "tf.random_shuffle"
print sess.run(tf_shuff)

# output(example):

# c
# [[1 2]
#  [3 4]
#  [5 6]]
# tf.random_shuffle
# [[5 6]
#  [3 4]
#  [1 2]]

######################
# tf.set_random_seed
######################
tf.set_random_seed(1234)
a = tf.random_uniform([1])
b = tf.random_normal([1])

print "tf.set_random_seed"
print "Session 1"
with tf.Session() as sess1:
    print sess1.run(a)
    print sess1.run(a)
    print sess1.run(b)
    print sess1.run(b)

print "Session 2"
with tf.Session() as sess2:
    print sess2.run(a)
    print sess2.run(a)
    print sess2.run(b)
    print sess2.run(b)

# output(example):

# tf.set_random_seed
# Session 1
# [ 0.81463408]
# [ 0.84950328]
# [-0.28481847]
# [ 0.23557714]
# Session 2
# [ 0.81463408]
# [ 0.84950328]
# [-0.28481847]
# [ 0.23557714]

sess.close()