[英]Tensorflow concat ragged tensor returning wrong shape
I need to concat two ragged tensors keeping the last dimension with a fixed size 2
.我需要连接两个参差不齐的张量,将最后一个维度保持为固定大小
2
。
Checking model.output_shape
I get the desired (None, None, 2)
.检查
model.output_shape
我得到了想要的(None, None, 2)
。 But when I call the model, I get (batch_size, None, None)
.但是当我调用模型时,我得到
(batch_size, None, None)
。 How do I get the right shape?如何获得正确的形状? Code:
代码:
import tensorflow as tf
a_input = tf.keras.layers.Input([None, 2], ragged=True)
b_input = tf.keras.layers.Input([None, 2], ragged=True)
output = tf.concat([a_input, b_input], axis=1)
model = tf.keras.Model([a_input, b_input], output)
a = tf.ragged.constant([
[[1, 2], [3, 4], [5, 6]],
[[1, 2], [3, 4]],
[[1, 2]],
])
b = tf.ragged.constant([
[[1, 2]],
[[1, 2], [3, 4], [5, 6], [7, 8]],
[[1, 2], [3, 4]],
])
print(model.output_shape)
# (None, None, 2)
print(model([a, b]).shape)
# (3, None, None)
I found it.我找到了。 The
tf.ragged.constant
does not consider the last dimension a uniform dimension. tf.ragged.constant
不认为最后一个维度是统一维度。 So a.shape
is (3, None, None)
.所以
a.shape
是(3, None, None)
。 To fix that I need to use ragged_rank
parameter:要解决这个问题,我需要使用
ragged_rank
参数:
a = tf.ragged.constant([
[[1, 2], [3, 4], [5, 6]],
[[1, 2], [3, 4]],
[[1, 2]],
], ragged_rank=1)
print(a.shape)
# (3, None, 2)
声明:本站的技术帖子网页,遵循CC BY-SA 4.0协议,如果您需要转载,请注明本站网址或者原文地址。任何问题请咨询:yoyou2525@163.com.