I am currently trying to solve a task with memoization. I have following recursion:
A (i, j) = f( A (i, j-1), A (i-1, j-1), A (i-1, j + 1) )
I am not sure in which order the sub-problems should be solved best, so that multiple calculations are avoided.
Does someone have an idea which order is best?
Let me show you the general approach to figuring this out. Then I'll let you apply that technique to your specific situation.
Well, it's clear that you need to solve $A(i,j-1)$, $A(i-1,j-1)$, and $A(i-1,j+1)$ before solving $A(i,j)$. So, you need to pick an order that ensures all three of those are solved before you try to solve $A(i,j)$. What order would work for that?
I suggest you try drawing a picture. Draw a grid, with one dot for each $(i,j)$. Draw an arrow from $(i,j-1)$ to $(i,j)$, an arrow from $(i-1,j-1)$ to $(i,j)$ and an arrow from $(i-1,j+1)$ to $(i,j)$ for each $(i,j)$. You might do this for, say, a 4x4 grid. Look at the resulting picture. Can you find an order of traversing the dots, so that is consistent with all of the arrows? Hint: try topological sorting that graph. What order does that give you? Can you generalize? What would work for a 5x5 grid?
If you're not sure, just write a recursive program and apply memoization. Then you won't need to explicitly pick an order to build things out bottom-up; the computer will take care of that for you. See When can I use dynamic programming to reduce the time complexity of my recursive algorithm? and https://en.wikipedia.org/wiki/Memoization.
