Hi, I have a question regarding the parameter “units” of SimpleRNN.
In the following function, what does the value “40” do?
tf.keras.layers.SimpleRNN(40, return_sequences=True)
Hi @heeseong_kim ,
it’s the number of units in this layer, in this case you have 40 SimpleRNN units that can be stacked with another layer, because you put return_sequences=True. You could do something like this:
tf.keras.layers.SimpleRNN(40, return_sequences=True)
tf.keras.layers.SimpleRNN(40, return_sequences=False)
in this case, the 2nd layer have a 40-dimensional sequence input, coming from the 1st layer.
Best
A little more expansive description is available at the RNN with Keras Guide in the section on Outputs and States, which reads in part…
By default, the output of a RNN layer contains a single vector per sample. This vector is the RNN cell output corresponding to the last timestep, containing information about the entire input sequence. The shape of this output is (batch_size, units) where units corresponds to the units argument passed to the layer’s constructor.
Could you elaborate on how the second line of your code fragment knows anything about the shape of the output of the first?
Yeah @ai_curious,
We are using the Sequential API, meaning the output of the 1st row simple_RNN is the input of the 2nd row simple_RNN. Take a look at this piece of model summary:
The classic RNN model has of a single hidden RNN layer, usually followed by a feedforward output layer. A Stacked RNN, like this one, is an extension to this model that has more hidden RNN layers where each layer contains multiple memory cells.
Best
1 Like
Thanks for the clarification. From the 2-line code fragment in the previous post, it didn’t seem to be using syntax I recognized as either Functional or Sequential.
You’re welcome,
and sorry for the misunderstanding: I’d skipped one comma and that could create some confusion.