Different kinds of layers Hyperparameter values

Are there any general guidelines based on what kind of data/ shape of the data/ sequence of layers on how to choose some of the below parameter values:

  • Number of units in hidden dense layers.
  • number of filters in a Conv2D layer
  • output_dim/embedding_dim in an embedding layer
  • the units/lstm_dim in an LSTM layer

We can test any number and experiment with what works best, but what is a good starting point? can we use some features of our data or the shape of the layers before it to choose these values?

Please see this topic.