问题描述
我已经运行 Keras LSTM演示包含以下代码(在第166行之后):
I have run a Keras LSTM demo containing the following code (after line 166):
m = 1
model=Sequential()
dim_in = m
dim_out = m
nb_units = 10
model.add(LSTM(input_shape=(None, dim_in),
return_sequences=True,
units=nb_units))
model.add(TimeDistributed(Dense(activation='linear', units=dim_out)))
model.compile(loss = 'mse', optimizer = 'rmsprop')
当我预先呼叫model.summary()时，会看到以下输出:
When I prepend a call to model.summary(), I see the following output:
_________________________________________________________________
Layer (type) Output Shape Param #
=================================================================
lstm_4 (LSTM) (None, None, 10) 480
_________________________________________________________________
time_distributed_4 (TimeDist (None, None, 1) 11
=================================================================
Total params: 491
Trainable params: 491
Non-trainable params: 0
我了解到时间分布层的11个参数仅由nb_units权重和一个偏差值组成.
I understand that the 11 params of the time distributed layer simply consist of nb_units weights plus one bias value.
Now for the LSTM layer:These answers say:
params = 4 * ((input_size + 1) * output_size + output_size^2)
在我使用input_size = 1和output_size = 1的情况下，这对于10个单位中的每一个仅产生12个参数，总计120个参数.与报告的480相比，误差减少了4倍.我的错误在哪里?
In my case with input_size = 1 and output_size = 1 this yields only 12 parameters for each of the 10 units, totaling to 120 parameters. Compared to the reported 480, this is off by a factor of 4. Where is my error?
推荐答案
params公式适用于整个图层，而不适用于每个Keras单位.
The params formula holds for the whole layer, not per Keras unit.
引用此答案:
Keras中的LSTM仅精确定义了一个LSTM块，其单元长度为单位长度.
LSTM in Keras only define exactly one LSTM block, whose cells is of unit-length.
直接设置output_size = 10(像这样评论)正确地产生了480个参数.
Directly setting output_size = 10 (like in this comment) correctly yields the 480 parameters.
这篇关于Keras:了解可训练的LSTM参数的数量的文章就介绍到这了，希望我们推荐的答案对大家有所帮助，也希望大家多多支持！