Sounds like someone I know ... lol
I've been involved with computers since the mid 60s, and have worked with many programming languages. I was self-taught in Assembler language on one of the earliest IBM 360's. It had 4 kilobytes of total memory, of which the operating system (then called the 'Supervisor') used 1K, the rest was available for the user. Needless to say, I learned to write very tight and efficient code. I also developed the early versions of two multi-$million software products.
Jai-alai was added to the mix in the late 70s, and I began collecting statistics and developing data models. Still do it. Unlike yourself, I have a heavy math stat background (MBA, Operations Research). Statistical confidence has become increasingly important as pool sizes shrink, for example, one of my inputs is the average payoff price for a specific combo. As the number of instances dwindles, the variance skyrockets.
Most of my jai-alai stuff is done in Visual Basic. Not the best for procedural coding, but I already had so much stuff in the can from my earlier PC work in BASIC that I stuck with it.
As you said, going from a player (or team) rating, regardless of how you get your rating, to a 'per point' rating for a Monte Carlo simulation is a huge issue. Anyway, after I adjust for a single point, I use a modified log 5 method.
My other major issue is trying to determine how heavily to rate more recent history in my exponential smoothing models. And, yes, I do subscribe to the 'When you're hot, you're hot' viewpoint. Most recently, I'm exploring the idea that each player's data might be smoothed with a different 'alpha'.
You and I, and perhaps 5 or 6 others that inhabit this board (or have in the past), are probably the only ones in the world who think about these things. Bring on the glazed-eye looks ... lol
