You shouldn't try to code a complete top-down solution without fully understanding the problem. (And how often do you fully understand the problem, or think that you do but don't?) But you can code up a bottom-up partial solution of the parts of the problem that you do fully understand, and in the process learn enough to make progress on the total solution, so long as the components you develop in the bottom-up process are composable and robust.
There was recently a thing I worked on where there were multiple similar things that needed to be done. Rather than finishing the beginning of A and then the beginning of B and then C, I opted to finish ALL of A then all of B then all of C. This way, what I learned when working on all of A, I could use to make better decisions about the beginning of B. Then what I learn from doing all of B would be used to help make the beginning of C even better.
If I did all the beginnings first, I'd then go back to doing the middle of A and realize I did the beginning wrong and have to go back to fix the beginnings of B and C.
