Number of inputs to LSTM

Course Q&A Deep Learning Specialization Sequence Models
1

While working on the Emojify V2, W2 assignment of course 5 of the DLS, I see that we are
asked to set 128 as the value for the input argument of the LSTM layer. What is the reason for choosing that number?

In fact, I don’t even understand what the argument input means. When I look at the architecture diagram, there are Tx (maxLen, I guess) LSTM cells. Please explain. Thank you!

2

Does this help?

3

Thank you! I had already read the API docs before posting the question. It did not help.

I found the following to be useful:

  1. python - How to interpret clearly the meaning of the units parameter in Keras? - Stack Overflow

  2. https://raw.githubusercontent.com/jasdeep06/jasdeep06.github.io/master/posts/Understanding-LSTM-in-Tensorflow-MNIST/images/num_units.png

Why number of units is 128 is still not clear? Why not 256 or 64 or 96?

4

You may try different units and see what the result/accuracy is. If you are told to use 128, that is based on experimentation which produces certain accuracy/result. To pass the test, you should follow the instructions.