TensorFlow函數(shù)：tf.matrix_transpose

tf.matrix_transpose 函數(shù)

matrix_transpose(
    a,
    name='matrix_transpose'
)

定義在：tensorflow/python/ops/array_ops.py.

參考指南：數(shù)學(xué)函數(shù)>矩陣數(shù)學(xué)函數(shù)

轉(zhuǎn)置張量 a 的最后兩個(gè)維數(shù).

例如：

# Matrix with no batch dimension.
# 'x' is [[1 2 3]
#         [4 5 6]]
tf.matrix_transpose(x) ==> [[1 4]
                                 [2 5]
                                 [3 6]]

# Matrix with two batch dimensions.
# x.shape is [1, 2, 3, 4]
# tf.matrix_transpose(x) is shape [1, 2, 4, 3]

注意,tf.matmul 提供 kwargs,它允許轉(zhuǎn)置參數(shù).這是以最小的成本完成的,并且比使用此函數(shù)更可取.例如：

# Good!  Transpose is taken at minimal additional cost.
tf.matmul(matrix, b, transpose_b=True)

# Inefficient!
tf.matmul(matrix, tf.matrix_transpose(b))

參數(shù)：

• a：張量,并且 rank >= 2.
• name：操作的名稱(可選).

返回值：

該函數(shù)返回一個(gè)經(jīng)過轉(zhuǎn)置的批次矩陣張量.

可能引發(fā)的異常：

• ValueError：如果 a 確定是靜態(tài)的,并且 rank < 2.