Function Differences with tf.nn.bias_add

tf.nn.bias_add

class tf.nn.bias_add(value, bias, data_format=None, name=None)

For more information, see tf.nn.bias_add.

mindspore.ops.bias_add

mindspore.ops.bias_add(input_x, bias)

For more information, see mindspore.ops.bias_add.

Differences

TensorFlow: Return the sum of the tensor of input value and bias, where bias is restricted to a 1D tensor and value supports various numbers of dimensions, and bias is broadcasted to be consistent with the shape of input value before the two are summed.

MindSpore: MindSpore API basically implements the same function as TensorFlow. However, MindSpore input input_x only supports 2-5 dimensional shapes.

Categories	Subcategories	TensorFlow	MindSpore	Differences
Parameters	Parameter 1	value	input_x	Same function, different parameter names
	Parameter 2	bias	bias	Same function
	Parameter 3	data_format	-	The data format of the input data. MindSpore does not have this parameter
	Parameter 4	name	-	Not involved

Code Example 1

The two APIs achieve the same function and have the same usage.

# TensorFlow
import tensorflow as tf

tf.compat.v1.disable_eager_execution()
value = tf.constant([[1, 2], [3, 4], [5, 6]], dtype=tf.float32)
bias = tf.constant([-2, -1], dtype=tf.float32)
result = tf.nn.bias_add(value, bias)
ss = tf.compat.v1.Session()
output = ss.run(result)
print(output)
# [[-1.  1.]
#  [ 1.  3.]
#  [ 3.  5.]]

# MindSpore
import numpy as np
import mindspore
import mindspore.ops as ops
from mindspore import Tensor

input_x = Tensor(np.array([[1, 2], [3, 4], [5, 6]]), mindspore.float32)
bias = Tensor(np.array([-2 , -1]), mindspore.float32)
output = ops.bias_add(input_x, bias)
print(output)
# [[-1.  1.]
#  [ 1.  3.]
#  [ 3.  5.]]