[英]How to declare multiple inputs LSTM model in Keras?
I have a Keras' code of declaring LSTM. 我有一个声明LSTM的Keras代码。 But I noticed Container class has already been removed in the latest version.
但是我注意到最新版本中已经删除了Container类。 https://keras.io/layers/containers/
https://keras.io/layers/containers/
How do I declare multiple inputs for LSTM in the latest format? 如何以最新格式声明LSTM的多个输入? I want to concatenate all inputs for the LSTM inputs.
我想连接LSTM输入的所有输入。
Although I noticed a similar post, what I want to do is declaration of the model. 虽然我注意到了类似的帖子,但我想要做的是声明模型。 How to work with multiple inputs for LSTM in Keras?
如何在Keras中使用LSTM的多个输入?
``` ```
g = Graph()
g.add_input(
name='i1',
input_shape=(None, i1_size)
)
g.add_input(
name='i2',
input_shape=(None, i2_size)
)
g.add_node(
LSTM(
n_hidden,
return_sequences=True,
activation='tanh'
),
name='h1',
inputs=[
'i1',
'i2'
]
)
``` ```
Oh, May I just set input_shape as (i1_size+i2_size) like below? 哦,我可以将input_shape设置为(i1_size + i2_size),如下所示吗?
model = Sequential()
model.add(LSTM(n_hidden, input_shape=(None, i1_size+i2_size), activation='tanh', return_sequences=True))
You asked: 您询问:
Oh, May I just set input_shape as (i1_size+i2_size) like below?
哦,我可以将input_shape设置为(i1_size + i2_size),如下所示吗?
model = Sequential() model.add(LSTM(n_hidden, input_shape=(None, i1_size+i2_size), activation='tanh', return_sequences=True))
Yes, Jef. 是的,杰夫。 Just keep in mind that your None in (None, i1_size+i2_size) is the number of RNN time steps/input_length and there are caveats to when you can skip defining it.
请记住,你的无(无,i1_size + i2_size)是RNN时间步数/ input_length的数量,当你可以跳过定义它时有一些警告。 Please see the description for
input_length
at https://keras.io/layers/recurrent/ for details. 有关详细信息,请参阅https://keras.io/layers/recurrent/上的
input_length
说明。
And just FYI input_shape=(None, i1_size+i2_size)
can also be written as input_dim=i1_size+i2_size
(assuming you don't include input_length
). 只有FYI
input_shape=(None, i1_size+i2_size)
也可以写成input_dim=i1_size+i2_size
(假设你不包含input_length
)。
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.