Using a neural network to convert numeral systems

Question

For starters. I know a neural.network is overkill for this but I'm doing it that way for the meme value. With that out of the way.

I'm trying to use a neural.network on either Tensorflow or Keras to solve the AoC day 25 2022.

The task is converting from integers to a made up numeral system and the other way around.

This is an excerpt from the "training data":

 Decimal          SNAFU
        1              1
        2              2
        3             1=
        4             1-
        5             10
        6             11
        7             12
        8             2=
        9             2-
       10             20
       15            1=0
       20            1-0
     2022         1=11-2
    12345        1-0---0
314159265  1121-1110-1=0

The task looks very similar to converting to and from Roman numerals but the only projects I've found doing that use CV and an image as input, not strings.

Honestly this is my first project using neural.network and I'd like to know where to start.

Any insights on what model should I use?

Answer 1

A sequence to sequence model is probably the best option. It can be done in Tensorflow using Attention:

Tensorflow attention tutorial

A very similar problem is translation to Roman numerals which is tackled in this example using MATLAB and the above approach