Shape of the inputs to the RNN

Course Q&A TensorFlow Developer Professional Sequences, Time Series and Prediction
1

Hello everyone,

I was taking a look at the following video:

It’s not clear to me why the output shape should be 4x3x30.
Particularly the “3” it’s unclear to me, he said that 3 is the number of neurons but I’m not able to get this part.

Are we saying that inside a cell there are 3 neurons?

Thank you very much in advance for the answer.
Manuel

2

Please move your topic to the correct subcategory.
Here’s the community user guide to get started.

See this to learn about units.

Deep learning specialization goes into the details of implementing an RNN from scratch.

Given that:

  • batch size = 4
  • number of timesteps = 30
  • number of features per timestep = 1
    the input shape you specify to the model is [30, 1] which excludes the batch dimension.

The model is configured with 3 units for the RNN cell. So:

  1. If return_sequences=True, output shape is (4, 30, 3).
  2. If not, output shape is (4, 3).

One rule of thumb is that when stacking RNN layers, use return_sequences=True. If not, you can leave the flag to False.

3

Thank you very much for your answer and sorry about the wrong location.
It’s now clearer to me!

Manuel