Computer Players

[dko]

Just curious, what average weighting do you have your computer player set at? Mine is at 81.1 and plays very erratically (kind of like me). All factors are set at 8 or higher (16+ for short iron and 24+ for putting.)

[Big Sexy JC]

DKO,

Here are the ranges that I use for golfer settings during my replays. I had 24 golfers per season so I had to have a range where the scores were fairly competitive, but the better golfers were better and the worse golfers were worse. The scores also had to be reflective of real PGA tour scores. After 44 seasons I have an enormous sample set of data to show these numbers are pretty spot on.

All golfers have a Pro range. My average score with these settings is about -2 with the worst player averaging about par and the best player averaging about -6.

Driver Range: 17-12.4
Long Irons Range: 8.5-6.2
Short Irons Range: 17-12.4
Fairway Woods Range: 8.5-6.2
Chipping Range: 8.5-6.2
Putting Range: 25.5-18.6

Total Golfer Score Range: 70 - Around 76ish (if their stats put them in to a Champ player I nerf them back down to a Pro level player)

[dko]

Thanks JC. I forgot that Driver is set at 2 points; my guy is at 15.6. He (it?) plays at Champ level which I can't change (as do I by choice) and I also use the classic swing. Your ranges are pretty broad so it's kind of impossible to draw a comparison on an individual player basis.

I was hoping to hear about the individual values used by folks who have one or a few AI's.

[cloeks1]

Check out the downloads at this location. I find these setting work very well and find it interesting how they were calculated.

viewtopic.php?f=8&t=1244

[dko]

Thank you. These ratings seem kind of low. I've never set an AI at lower than 80, and while I consider myself an average player, I usually play pretty even with it.

I seem to remember someone else set ratings for the top players (Arnold, Jack, etc.) and if memory serves they were over 80.

Tiger Woods is 66.5? Maybe in 2021, but come on,,,

[Adelade]

A few years ago I made some computer players to play with and against in Ryder Cup simulations at Le Golf National and tweaked them to score similarly to real life, Im not saying I managed to tweak them just right, because I only played with and against computer players for a short while, I never do it nowadays, but I think I was reasonably happy with them, they ended up being between 77.7 and 79.0.

It is probably worth noting that their scores ought to depend on difficulty settings like green speeds, tee and pin choices, wind, type of course etc. I would be curious to hear which settings you normally used Big Sexy JC.

[Big Sexy JC]

Adelade,

I feel like we might be headed down the proverbial rabbit hole with this topic, but here goes. I'll try to keep this simple but after setting up 1056 AI golfers for my replay I could probably write a short book on this.

The easy answer to your question is that my greens were always setup as Medium and Moderate (If I had another year of time to redo my replay I would've gone with Challenging Medium and Challenging Moderate instead). I always played with no wind. As for pin placements, the short and easy answer is I used custom/specific difficult pins.

However, I tracked the data for every pin placement for every course and the resulting score from all 24 golfers for that course and tried to use pin placements that created realistic scores. That means that after I gathered enough data, I used Moderate on some courses, Easy on some courses and Difficult on most courses. In every case though I used specific pin placements or a range of specific pin placements to get the scores I was after.

Here is my .02 on AI golfers as well. What determines if an AI golfer is rated as Pro or Champ is 100% based on the algorithm the game uses to decide that. Before I started my replay I tested the absolute crap out of that algorithm and pretty much know how it works now. I found the Pro rated golfers to be the most realistic. The Champ golfers generally scored too low because if the algorithm deemed them to be Champ then that meant their ratings were so good in enough categories that nothing was going to keep them from scoring well.

If you want AI golfers that score similar to real life PGA players then I have that down to a science. If you want AI golfers that compete evenly with your scores or skill then that will take some testing. I can tell you that golfers that are in the 75 to 77 range and still rated as Pro are going to average around -4 to -6 per round over the long haul. Obviously, conditions could cause some variance to that.

Below is a link to every AI golfer I've used so feel free to use them, test with them, etc..

https://drive.google.com/drive/folders/ ... sp=sharing

Below is a link to all the spreadsheets with the yearly and career stats.

https://drive.google.com/drive/folders/ ... sp=sharing

Below is a link to the individual year spreadsheets.

https://drive.google.com/drive/folders/ ... sp=sharing

[Adelade]

Good stuff. I thought more about it, and perhaps there is a chance that the scoring has nothing to do with the conditions after all. I remember the computer players completely sandbagging some holes here and there, which always made me suspect that they operate under some type of target score, and whenever the computer player realizes that it is on the way to scoring a bit too well, it plays horrible for a hole to make sure of a bogey or so. Many times they just hit an absolutely failed drive hooking almost 45 degrees to the left and only carrying like 100 yards, a shot that might be realistic for 20+ handicappers, but a pro golfer? No way. Similarly, if they have a 30-feet putt they sometimes only hit it 15 feet or so, then the next too short too, seemingly to make sure of a 3-putt.... makes no sense at all. Then next hole they play comepletely fine again...

[dko]

And down the rabbit hole we go (my bad for innocently opening this topic.)

I only play on the actual courses used on the PGA Tour using pin positions and hole yardages based on actual tournament play from 2016 to 2023. I play at Medium/Fast Challenging under either breezy or gusty conditions based on the actual weather and play four rounds on each course/year. I know, BORING (except to me.)

As my AI progressed from Pro to Champ (I wasn't aware of this algorithm so don't know when it happened) I didn't notice much difference in scoring; the AI's scoring averaged from 0.4 over par to 2.1 under par over the years (no discernible trend). The AI was always set at an average skill level of 81.1.

It did poorly at Augusta, Bay Hill, East Lake and Muirfield Village and very well on Waialae and most of the TPC courses. I did much better on Bay Hill and worse on Waialae, but otherwise had similar results.

So, there could be something about the impact of courses and conditions selected.

And yes, the AI does have the one-off hole that sometimes derails its round (this is particularly true at hole 6 on Bay Hill) that results in double bogey or worse (such as the dreaded pickup 12.) While I have not seen the 45-degree drive that travels 100 yards (unless a tree gets in the way) or the putt that stops 15 feet short, there has occasionally been the use of a 3-wood from a greenside bunker and other odd club selections. It also misses a lot of short putts when there is a bit of a break involved, but chips in more frequently than I like.

[Big Sexy JC]

Doug,

A couple of things. First, I'm totally with you on the PGA Tour courses. As I've mentioned in other posts, I have a real life simulation addiction going back over two decades so I find absolutely no value in fictional/fantasy courses. I tried to play every replay with the exact courses the tour used and tried to keep them in the same order as if I were trying to prevent my AI players from getting digital jet lag.

Second, if an AI player has an 81.1 average then he will always be a Champ player. The highest score the algorithm will allow (that I was able to discover) for a Pro player is 78.8 and the lowest score for a Champ player is roughly 75.5. I'll give those examples below. So you can technically have a Pro player with a higher score than a Champ player, but that's because the algorithm weights the categories differently.

Part of the weighting is obvious with the categories that are able to score higher (driving, short irons and putting). However, there are also thresholds around 16 for driver and short irons and 8 for chipping, long irons and fairway woods and around 24 for putting. For example, once you rate a player above a 16 in driving you must keep the long iron, fairway wood and chipping below 8 or you will push the player into the Champ level. How high the putting score is also factors into the equation. You can have a high putter and driving score, but the rest of the numbers need to be lowered to keep the player at a Pro level. I could go on forever with examples, so I'll share just a few below. You can test these with your game. I have a spreadsheet with all the different possibilities and what level it produced that I created when I was testing the algorithm.

I pulled the examples below from my spreadsheet as the most extreme I was able to find in my testing. Again, notice the thresholds and the number of categories that are at those thresholds and the corresponding level of the player.

Points Level - Pro
Tee 14.8
Long Iron 7.9
Short Iron 14.8
Fairway Wood 7.9
Chipping 7.9
Putting 25.5
78.8

Points Level - Champ
Tee 12.4
Long Iron 8
Short Iron 16
Fairway Wood 8
Chipping 8
Putting 23.1
75.5


Points Level - Pro
Tee 15.6
Long Iron 7.9
Short Iron 15.8
Fairway Wood 7.9
Chipping 7.9
Putting 23.4
78.5

Points Level - Pro
Tee 15.8
Long Iron 7.9
Short Iron 15.8
Fairway Wood 7.9
Chipping 7.9
Putting 23.1
78.4

Points Level - Pro
Tee 16.8
Long Iron 8.5
Short Iron 16.8
Fairway Wood 8.5
Chipping 8.5
Putting 18.6
77.7

Points Level - Champ
Tee 17
Long Iron 6.2
Short Iron 15
Fairway Wood 6.2
Chipping 6.2
Putting 25.5
76.1

Points Level - Champ
Tee 17
Long Iron 7
Short Iron 12.4
Fairway Wood 7
Chipping 7
Putting 25.5
75.9

I am 100% positive that the conditions factor in to the AI scoring. I have seen some weird stuff that you and Adelade mention, but I don't think the AI scores are scripted, but there is some random variance that can lead to some interesting scoring. One of the biggest factors I've seen in my replays that impact scoring is how well the course is laid out. For example, there are probably 6-10 courses that a human can navigate without issue so during play testing it was never discovered that an AI player has no idea how to navigate some of the holes the way they are laid out. Bay Hill has at least 1, the Augusta before the new release on hole 13 was a disaster and TPC Southwind and Muirfield Country Club both have at least 1 hole that destroys AI scoring.

The ratings are the biggest factor though, and they should be. In my seasons, I take the top 8 scorers and have a March Madness type Match tourney to determine my season champ. So 8 make the tourney and 16 dont. In 44 seasons there have only been 3 players to make the tourney with a score under 74.5. In every case, those players had very high driving ratings and were able to score well with weaker putting, long iron and fairway wood ratings. The game tends to favor the long drive AI players over the great putters, which is why if I could do my replay over again I would make putting more difficult to level the playing field.

I love that you started this topic Doug. I enjoy making anis and clubs and apparel, but I am an absolute numbers nerd so I could chat about this stuff for days. The only thing that has kept Links playable for me was the fact that I could create AI players with customizable ratings. No other game has done it and I don't see another one doing it in the future.