tf.reshape()

1.tf.reshape

?

1	reshape(tensor, shape, name=None)

作用：重塑张量。给定张量，此操作将返回与形状为shape的张量具有相同值的张量。 如果“形状”的一个分量为特殊值-1，则将计算该尺寸的大小，以使总大小保持恒定。 具体来说，[-1]的“形状”会展平为一维。 “形状”的至多一个分量可以为-1。 如果“ shape”为一维或更高，则该操作将返回一个形状为“ shape”的张量，其中填充了“ tensor”的值。 在这种情况下，“形状”所隐含的元素数量必须与“张量”中的元素数量相同。 

举例：

+ View Code?

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54

For example:   ```   # tensor 't' is [1, 2, 3, 4, 5, 6, 7, 8, 9]   # tensor 't' has shape [9]   reshape(t, [3, 3]) ==> [[1, 2, 3],                           [4, 5, 6],                           [7, 8, 9]]     # tensor 't' is [[[1, 1], [2, 2]],   #                [[3, 3], [4, 4]]]   # tensor 't' has shape [2, 2, 2]   reshape(t, [2, 4]) ==> [[1, 1, 2, 2],                           [3, 3, 4, 4]]     # tensor 't' is [[[1, 1, 1],   #                 [2, 2, 2]],   #                [[3, 3, 3],   #                 [4, 4, 4]],   #                [[5, 5, 5],   #                 [6, 6, 6]]]   # tensor 't' has shape [3, 2, 3]   # pass '[-1]' to flatten 't'   reshape(t, [-1]) ==> [1, 1, 1, 2, 2, 2, 3, 3, 3, 4, 4, 4, 5, 5, 5, 6, 6, 6]     # -1 can also be used to infer the shape   # -1 is inferred to be 9:   reshape(t, [2, -1]) ==> [[1, 1, 1, 2, 2, 2, 3, 3, 3],                            [4, 4, 4, 5, 5, 5, 6, 6, 6]]     # -1 is inferred to be 2:   reshape(t, [-1, 9]) ==> [[1, 1, 1, 2, 2, 2, 3, 3, 3],                            [4, 4, 4, 5, 5, 5, 6, 6, 6]]     # -1 is inferred to be 3:   reshape(t, [ 2, -1, 3]) ==> [[[1, 1, 1],                                 [2, 2, 2],                                 [3, 3, 3]],                                [[4, 4, 4],                                 [5, 5, 5],                                 [6, 6, 6]]]     # tensor 't' is [7]   # shape `[]` reshapes to a scalar   reshape(t, []) ==> 7   ```   Args:     tensor: A `Tensor`.     shape: A `Tensor`. Must be one of the following types: `int32`, `int64`.       Defines the shape of the output tensor.     name: A name for the operation (optional).     Returns:     A `Tensor`. Has the same type as `tensor`.   """

bert中源码：　　

+ View Code?

1 2 3 4 5 6 7 8 9 10 11

# If the input is a 2D tensor of shape [batch_size, seq_length], we # reshape to [batch_size, seq_length, 1]. if input_ids.shape.ndims == 2:   input_ids = tf.expand_dims(input_ids, axis=[-1])   embedding_table = tf.get_variable(       name=word_embedding_name,       shape=[vocab_size, embedding_size],       initializer=create_initializer(initializer_range))   flat_input_ids = tf.reshape(input_ids, [-1]) #【batch_size*seq_length*input_num】