New Baseball Sim Feedback

[pure]

Hi folks, after spending the last few years building the "Au" engine that powers the UI for all the recent Wolverine Studios sims, I am working on building a Baseball game of my own on the engine.

Having worked in machine learning for the last ten years, I am looking at novel ideas that employ ML behind the engine (in particular for GM and Manager AI)

I'd like to post here to get feedback along the way; and today I have the first run of the ratings generation code for hitters. I think in its current state it may penalize players that had very few appearances a little too much, but I'd love your feedback.

Model Features - Test Run #1 - Batters - Google Sheets

[pure]

Here is a quick peek at the Physics model, I simmed 300 PA for a batter with a POWER rating of 80.

The engine calculates launch angle and exit velocity for all hits.

Here are the HR for that sim session:

Statcast Model Test - EV and LA - Google Sheets

[21C]

Looking forward to following your progress. Good luck with it, Shaun.

[Atocep]

Launch angle and exit velo data looks solid. An 80 power per your model would be a Mookie Betts type player and those exit velos look in line with his numbers. The sweet spot for homers is 24 to 32 degree launch angle and most of these homers fall in that range.

Just a quick look, but I'd say you're definitely on the right track.

[JPhillips]

It seems like you have too many 95+ players. If you max out at 100 you need to allow space for historical players to be great.

[Ben E Lou]

Originally Posted by JPhillips

It seems like you have too many 95+ players. If you max out at 100 you need to allow space for historical players to be great.

This.

[pure]

I'm not sure if its clear, but the ratings in that sheet are relative to one season (e.g. if you started a replay in 1972, what would the player's initial, unadjusted ratings be relative to others in that season).

Given that, do you still think there are too many high 90's? Maybe I need to play with the curve. If you plot it now, it would be close to a normal distribution for each rating.

[BYU 14]

Oh dang, a Shaun Sullivan sighting? Looking forward to anything you put out, will be following.

[JPhillips]

Originally Posted by pure

I'm not sure if its clear, but the ratings in that sheet are relative to one season (e.g. if you started a replay in 1972, what would the player's initial, unadjusted ratings be relative to others in that season).

Given that, do you still think there are too many high 90's? Maybe I need to play with the curve. If you plot it now, it would be close to a normal distribution for each rating.

Let me start by saying you know what you want to do much better than I do, and you've got the programming expertise that I don't.

But, I do think baseball is defined by its history in a way that other sports aren't. We're always comparing players to players from the past. If multiple players each season are the best possible, that would make those comparisons less effective for me.