First, prove that it’s possible

Building on my previous post about finding a working solution first, before thinking of optimization issues.

I have been fortunate to work on many interesting problems in my career, and I have used that approach every time: first, find something that works. Maybe it works with a subset of the actual data, or solves the problem for specific cases. Maybe the initial solution takes two full days to come up with a solution using only 0.1% of the data. Or perhaps it’s only a partial solution.

That’s okay! As they say, Rome wasn’t built in a day.

Software development involves a certain amount of discovery: figuring out how to do things. The first part of figuring out how to do something is determining if it’s even possible. You need an existence proof. Once you have an existence proof, everything else is just optimization.

Don’t laugh. I’m not trying to minimize the amount of work that might be involved in optimizing a general solution. Rather, I’m saying that no amount of optimization can possibly help if you don’t know whether a solution is possible.

As a senior developer, I’ve dealt with this too many times. Management decides they want a new feature and immediately builds a team to start implementing it, not stopping to ask themselves whether a solution is possible. In one memorable case the problem they wanted to solve was essentially a 10,000 node traveling salesman problem to which they wanted the optimum solution. They built a team with a half dozen programmers and set them to work building the backend infrastructure and asked me to lead the development effort. It didn’t take me very long to determine that what they wanted to do was impossible. It took months to convince them that we could provide a “good enough” solution (at a huge cost, I might add), but that we couldn’t provide a guaranteed optimum solution at any price.

Programmers often aren’t any better. Faced with a problem, many programmers immediately begin searching their brains for all the optimization tricks they’ve learned over the years. That’s all well and good, except they often forget the first step: is a solution even possible.

Upon reflection, that is one of the key differences between a computer science curriculum and a programming job. An undergraduate computer science curriculum teaches one how to solve problems that have known solutions. The basic assumption behind every assignment is that a solution does in fact exist. The student’s task is to discover and implement that solution. But in industry we are often faced with problems for which there is no known solution. We have to determine whether a solution is even possible.

I’ll admit that, even now, I usually approach new problems with the assumption that a solution is possible. But, having been around the block a time or three, I’m prepared for the possibility that a solution is not possible, or not possible given the constraints under which I am working. That’s where things get really interesting. Anybody can write a program to find “the answer.” It takes an entirely different set of skills to write a program that selects the “best” or “good enough” answer within the constraints of time and budget.