Python – Tensorflow math.accumulate_n() method

Tensorflow math.accumulate_n() method performs  the element-wise sum of a list of passed tensors. The result is a tensor after performing the operation. The operation is done on the representation of a and b. This method belongs to math module.

Syntax: tf.math.accumulate_n( inputs, shape=None, tensor_dtype=None, name=None) Arguments

• inputs: This parameter takes a list of Tensor objects, and each of them with same shape and type.
• shape: This is optional parameter and it defines the expected shape of elements of inputs.
• dtype: This is optional parameter and it defines the expected data type of inputs.
• name: This is optional parameter and this is the name of the operation.

Return: It returns a Tensor having the same shape and type as the elements of inputs.

Let’s see this concept with the help of few examples: Example 1:

Python3

# Importing the Tensorflow library
import tensorflow as tf

# A constant a and b
a = tf.constant([[1, 3], [6, 7]])
b = tf.constant([[5, 2], [3, 8]])

# Applying the accumulate_n() function
# storing the result in 'c'
c = tf.math.accumulate_n([a, b, b])

# Initiating a Tensorflow session
with tf.Session() as sess:
    print("Input 1", a)
    print(sess.run(a))
    print("Input 2", b)
    print(sess.run(b))
    print("Output: ", c)
    print(sess.run(c))

Output:

Input 1 Tensor("Const_67:0", shape=(2, 2), dtype=int32)
[[1 3]
 [6 7]]
Input 2 Tensor("Const_68:0", shape=(2, 2), dtype=int32)
[[5 2]
 [3 8]]
Output:  Tensor("AccumulateNV2_2:0", shape=(2, 2), dtype=int32)
[[11  7]
 [12 23]]

Example 2:

Python3

# Importing the Tensorflow library
import tensorflow as tf

# A constant a and b
a = tf.constant([[2, 4], [1, 3]])
b = tf.constant([[5, 3], [4, 6]])

# Applying the accumulate_n() function
# storing the result in 'c'
c = tf.math.accumulate_n([b, a, b], shape =[2, 2], tensor_dtype = tf.int32)

# Initiating a Tensorflow session
with tf.Session() as sess:
    print("Input 1", a)
    print(sess.run(a))
    print("Input 2", b)
    print(sess.run(b))
    print("Output: ", c)
    print(sess.run(c))

Output:

Input 1 Tensor("Const_73:0", shape=(2, 2), dtype=int32)
[[2 4]
 [1 3]]
Input 2 Tensor("Const_74:0", shape=(2, 2), dtype=int32)
[[5 3]
 [4 6]]
Output:  Tensor("AccumulateNV2_5:0", shape=(2, 2), dtype=int32)
[[12 10]
 [ 9 15]]