如何在Tensorflow 2.0中创建填充批处理？

Question

我正在尝试从tensorflow数据集创建填充批处理，这使我出错。

样本数据：

A = [
 ([[101,9385,13302,102], [1,1,1,1], [0,0,0,0]]),
 ([[101,9385,13302], [1,1,1], [0,0,0]]),

 ([[101,9385,13302,102], [1,1,1,1], [0,0,0,0]]),
 ([[101,9385,13302], [1,1,1], [0,0,0]])     
 ]

预期输出：

Output = [
 ([[101,9385,13302,102], [1,1,1,1], [0,0,0,0]]),
 ([[101,9385,13302,-100], [1,1,1,-100], [0,0,0,-100]]),

 ([[101,9385,13302,102], [1,1,1,1], [0,0,0,0]]),
 ([[101,9385,13302,-100], [1,1,1,-100], [0,0,0,-100]])     
 ]

下面是代码：

chk = tf.data.Dataset.from_generator(lambda: A,
                                         output_types = (tf.int32))
BATCH_SIZE = 2
chk_batched = chk.padded_batch(BATCH_SIZE ,
                                   padded_shapes=(4),
                                   padding_values=(0))

for elem in chk_batched.as_numpy_iterator():
   print(elem)

错误在下面给出：

InvalidArgumentError: All elements in a batch must have the same rank as the padded shape for component0: expected rank 1 but got element with rank 2

Answer 1

因此它是二维的，papped_shapes =（4,3），无需编写padding_values =（0）。默认情况下，填充的值为0