Function Differences with tf.eye

tf.eye

tf.eye(
    num_rows,
    num_columns=None,
    batch_shape=None,
    dtype=tf.dtypes.float32,
    name=None
) -> Tensor

For more information, see tf.eye.

mindspore.ops.eye

mindspore.ops.eye(n, m, t) -> Tensor

For more information, see mindspore.ops.eye.

Differences

TensorFlow: It is possible to accept batch_shape in the parameters in TensorFlow to make the output have such a shape.

MindSpore: The number of columns and data types cannot be defaulted, and there is no difference in function.

Categories	Subcategories	TensorFlow	MindSpore	Differences
Parameters	Parameter 1	num_rows	n	Same function, different parameter names
	Parameter 2	num_columns	m	Specifies the number of columns of the tensor. Optional in TensorFlow; without this parameter, a tensor with the same number of columns and rows is returned; required in MindSpore
	Parameter 3	batch_shape	-	Makes the output have the specified shape. MindSpore does not have this parameter. For example, `batch_shape=[3]`
	Parameter 4	dtype	t	The name is different, optional in TensorFlow. If not, the default is `tf.dtypes.float32`; required in MindSpore
	Parameter 5	name	-	Not involved

Differences Analysis and Examples

Code Example 1

TensorFlow can default num_columns, and MindSpore cannot default.

# TensorFlow
import tensorflow as tf

e1 = tf.eye(3)
print(e1.numpy())
# [[1. 0. 0.]
#  [0. 1. 0.]
#  [0. 0. 1.]]

# MindSpore
import mindspore
import mindspore.ops as ops
e1 = ops.eye(3, 3, mindspore.float32)
print(e1.numpy())
# [[1. 0. 0.]
#  [0. 1. 0.]
#  [0. 0. 1.]]

Code Example 2

TensorFlow can default dtype, and MindSpore cannot default.

# TensorFlow
import tensorflow as tf
e2 = tf.eye(3, 2)
print(e2.numpy())
# [[1. 0.]
#  [0. 1.]
#  [0. 0.]]

# MindSpore
import mindspore
import mindspore.ops as ops
e2 = ops.eye(3, 2, mindspore.float32)
print(e2)
# [[1. 0.]
#  [0. 1.]
#  [0. 0.]]