Making a network with a variable number of inputs

I want a NN that can process a list of objects that with variable length and point at one of the objects. Similar to this: https://arxiv.org/pdf/1511.06391.pdf

The problem is, that the objects in the list are choices in a game and consist of two parts that I want to preprocess in the network. The list can be anything from 2 to 20 choices long and in some cases it might even get longer, so I don't know it's length in advance.

A way I could do this would be something like

x = []  for thing in list:    x.append(preprocess_network(thing))  output = pointer_network(x)  

though I don't know how to train network1 in this case.

https://stackoverflow.com/questions/67363352/making-a-network-with-a-variable-number-of-inputs May 03, 2021 at 12:06PM