STHS Forum

Hi everyone.

As you might have guested from my last message, I have already started programming the step #2 (passing, scoring and goalies) of version 2.1 of simulation engine. Actually, I did a lot of try and error on the Valentine’s Day week-end and I got some good results at the end of it. So last week, I had fun running A LOT of automatic test. I ran automatic tests with 5 different ratings pack on the 2 simulation engines (V20 and V21). Each Automatic Test had run with the following game setting.

Fight: 50
Hits: 45 and 55
Penalty: 44 and 55
Shots: 30,40,50,60 and 70
Goals: 30,40,50,60 and 70

The following Rating packs were used: ACCH2006-2007, ACCH-2007-2008, ACCH-2008-2009, ACCH-2009-2010 and a pack called SameRating where all forward had the same rating, all defenseman had the same rating and all goalies had same rating.

So after 1000 season of 1350 match, I have a LOT of data! And today, I’m publishing this data! I’m providing you with 5 download links because they are a big download (Around 35 MB each). Each download will have the following data for V20 and V21:

• A fusion of the goalies CSV file into one big Excel file format by myself.
• A fusion of the players CSV file into one big Excel file format by myself
• A fusion of the teams CSV file into one big CSV file.
• The AutomaticTest.csv file

Here are the download links:

http://sths.simont.info/DownloadFiles.p ... CH0607.zip
http://sths.simont.info/DownloadFiles.p ... CH0708.zip
http://sths.simont.info/DownloadFiles.p ... CH0809.zip
http://sths.simont.info/DownloadFiles.p ... CH0910.zip
http://sths.simont.info/DownloadFiles.p ... Rating.zip

Have fun!

it seems like V20 was better...I dunno why but specially Mike Fisher and Vinny Prospal seem really dominant

I did a lot of try and error on the Valentine’s Day week-end

so what you are saying is ......if you were not single before Valentine's day
you are single now ?

No, sadly for me *because she is reading my type right now*, I'm still in a couple.

I did some test and I tried to adjust ratings and settings to get realistic results and I had some difficulties. One thing I noticed, goalies are really awful in 2nd and 3rd period. Typically, there was between 7 and 9 goals per game (2 teams together), but sometimes there was over 20 goals. In one of those game, Ryan Miller, a top 5 goalie in the league, allowed 24 goals on 25 shots…
He started the game, allowed a goal on the first shot, stopped the 2nd shot, then allowed 3 goals on the next 3 shots. The backup came in the game, allowed 4 goals on 10 shots. Ryan Miller returned to the game and allowed 20 goals on 20 shots. That's the worst game I saw (not on total goals but for a star goalie like Miller)

In another game, Antti Niemi replaced Biron in the 3rd period, he stopped the first 4 shots and after that he did not save a puck. 9 shots. 9 goals.



I analysed a bit more the stats and to confirm what I saw in my test, I checked the data in your ACCH 09-10 test. I checked the distribution of goals in the 3 periods (I excluded overtime)

In your test, (all the settings produced similar results)
1st period: 4.5 to 8.37% of goals
2nd period: 34.9 to 46.9% of goals
3rd period: 46.9 to 60.3% of goals

Shots are much more balanced, but still the 1st period is behind the 2 others (about 23% of shots in the 1st, 38% in each of 2nd and 3rd period)

I tested with or without fatigue, the distribution is the same. I tested with fatigue settings so high that it was not a factor and still the same results) (players would have to play 90 minutes and goalie receive 90 shots to be fatigued) So the problem is somewhere else…

I can confirm these findings as I have seen similar distribution in my tests.

bond007 wrote:In your test, (all the settings produced similar results)
1st period: 4.5 to 8.37% of goals
2nd period: 34.9 to 46.9% of goals
3rd period: 46.9 to 60.3% of goals

As I mention in this post: http://sths.simont.info/Forum/viewtopic.php?f=29&t=8266, I know why and I'll be fix in 2.0.5. Actually, I had completely redid the goal formula in 2.0.5 for V21 engine.

I still have some games with ridiculously high score with v2.0.5. It's better than v2.0.4 because of the 2nd and 3rd period but there is something wrong.
It's concentrated against 3 teams...

Devils (Dwayne Roloson, average #1 goalie, save percentage 0f 0.010, GAA 21.37, 0 wins 74 loss)
Thrashers (Ryan Miller, ratings over 95 everywhere, save percentage of 0.016, GAA of 22.22, 0 wins, 64 loss)
Blue Jackets (Marty Turco, save pct 0.012, GAA of 22.10, 0 wins 70 loss)

Their backup have normal stats. I traded the goalies to another team and the problem is transfered to the other team
so there is something about those goalies that the simulator doesn't like.

Here is one exhibition game between 2 of those teams
http://www.lchv.biz/LCHV9-Exh247.html

and the rosters
http://www.lchv.biz/LCHV9-ProTeamRoster.html

Send me your league file so I can check it.

Check this : http://sths.simont.info/Forum/viewtopic ... 289#p35289

Thank you, the goalies are ok.


Is it normal that with, for example, ACCH 2009-2010 ratings and an automatic fantasy draft or real NHL roster, with
v2.0.5, after I adjusted sliders (shots 41, goals 53, hits 60, fight 35, Penalty 30) to get league average near what I'm looking for,
2.9 goals/game
21% PP
30 shots/game
14 PIM/game
21.7 hits/game

The best players get over 100 goals and 200 points with a shooting % over 30% in 82 games.
The best one, Zach Parise, finished with 148 goals, 161 pass, 48.8 shoot %.

I have 1 player over 300 points, 3 between 200-299, 5 between 150 and 199 and 31 between 100 and 149. Of course to keep the league average normal, the 3rd and 4th line players doesn't get many points. I was not expecting such a difference.

Is it the normal behavior? We have to reduce considerably difference in the ratings?

That behavior is normal. I change the formula because of the conclusion I did in this topic :h ttp://sths.simont.info/Forum/viewtopic.php?f=29&t=8233#p35087

My conclusion: In the NHL, goal scoring over a huge sample do a logarithm function (For a sample of 583 players and on a scale from 0 to 50 players, only 25 players have 25 or more goals) while in the STHS, the formula is more linear. I think that why you guy complain a lot on that. This will give me food for thought.

So I work hard to making sure players with very SC score a lot and that the league stat became a logarithm and not linear.

I don't know how STHS calculate probability to score but in my mind, it should be close to linear

Of course there are very few player scoring 25 goals or more. But there are also very few players with high SC rating.
For example If there is 25 players over 80 SC and 100 between 60 and 80SC, the best ones can get 40-50 goals while players with 65-70 SC get 15-20 goals and the relation between goals and SC would be near linear. And we didn't even talk about the fact that best players get more quality ice time in PP which increase their stats.


The only way I have been able to get statistics that were not too bad was giving almost the same rating to everyone (for example SC, PA, DF,... from 60 to 80. I did the same thing for goalies too).
It's sad we can only use about 20 numbers on the 0-100 scale STHS allow us to use without having "extreme" results.

For the moment I didn't really check if those results were realistic (if the best players got the best stats most of the time, if there was some randomness,...). I was only trying to get statistics near what it should be.

What I did is list all players goal in Excel and put the dat in a graph. It's cleary show a logarithm graph and not a linear graph.

Because your graph make a relation between goals and... what? His rank in the NHL?

Then yes, it's not linear.

But if's in relation with talent... it's completely different because there are many "average" players but only few exceptional players. Of course in the NHL it's hard to make a relation between goals and something because talent is not really measured. But with STHS, it is. Except if the players were equally distributed (5 players with 99, 5 with 98, 5 with 97,...)... if it's not linear, it would be close to.