Python 如何将RaggedSensor与tf.data API一起使用
我想将RaggedSensor与tf.data API一起使用,但似乎张量形状的计算不正确。 下面是一个将批次转换为RaggedSensor并返回张量的示例:Python 如何将RaggedSensor与tf.data API一起使用,python,tensorflow,tensorflow-estimator,Python,Tensorflow,Tensorflow Estimator,我想将RaggedSensor与tf.data API一起使用,但似乎张量形状的计算不正确。 下面是一个将批次转换为RaggedSensor并返回张量的示例: import tensorflow as tf shapes = [None] types = tf.string defaults = "<pad>" def generator_fn_ragged(): yield ( ["The", "brown", "fox", "jumps"],
import tensorflow as tf
shapes = [None]
types = tf.string
defaults = "<pad>"
def generator_fn_ragged():
yield (
["The", "brown", "fox", "jumps"],
["The", "brown", "fox", "jumps", "over", "the", "lazy", "dog"],
)
def input_fn():
dataset = tf.data.Dataset.from_generator(
generator_fn_ragged,
output_shapes=shapes,
output_types=tf.string,
)
dataset = dataset.padded_batch(2, shapes, defaults)
def _ragged(*features):
return [tf.RaggedTensor.from_tensor(x, padding="<pad>") for x in features]
dataset = dataset.map(_ragged)
def _unragged(*features):
return [x.to_tensor(default_value="<pad>") for x in features]
dataset = dataset.map(_unragged)
return dataset
dataset = input_fn()
for el in dataset:
print(el)
break
你确定这里必须使用不规则的张量吗?你到底想实现什么?你确定你必须在这里使用参差不齐的张量吗?你到底想达到什么目的?
(<tf.Tensor 'IteratorGetNext_7608:0' shape=<unknown> dtype=string>,)
Tensor("IteratorGetNext_4:0", shape=(?, ?), dtype=string)