Reliever's Appearances way too high! (Full Version)

All Forums >> [Current Games From Matrix.] >> [Sports] >> PureSim Baseball >> Tech Support



Message

sposfan -> Reliever's Appearances way too high! (11/23/2006 6:55:14 AM)

My main association (1982-1988 so far), always using latest builds (currently 1.40). Checked the regualr almanac for single season records and they are all within reason and realism EXCEPT for appearances where the top 10 are all over 100 appearances (MLB record is Mike Marshall with 106 and only 3 pitchers have recorded 90+ with Marshall doing it 3 times, Tekulve doing it 3 times as well (including the latest in 87) and Wayne Granger hitting 90 once.

In mine the top 10 have: 135, 126,123,121,117,116,112,111,109 and 107.

This is playing with the default XML and real players. One thing I have noticed is that all but one of the top 10 records were set in the past 2 years of my association.

107 and 109 I can live with, but 135 out of 162 games??!?! It seems that the CM really rides the relievers in the #1 slot extremely hard.

Upon further review in the HTML almanac the top 100 for single season appearances goes down to 82, meaning 100 pitchers have appeared in more than one half of their team games compared to MLB where 50 players have appeared in 82 plus games, but that's in a span of 40 years ( first one in 1965 and the last in 2005). At this rate if I manage to get 40 years in hundreds of pitchers will have eclipsed this mark.

Now, this is not a major issue in my eyes and I can probably tweak the XML easily enough to bring appearances into line, but it would be nice if it worked out of the box. The main thing is that ALL of the other stats line up extremely well so kudos to Shaun if the only real problem I found is that relievers appear in too many games!



puresimmer -> RE: Reliever's Appearances way too high! (11/23/2006 4:43:28 PM)

I'll look into that. There has been a long standing stuggle with reliever usage in the AI...

I think I can make a few tweaks here. Are you using real players or fictionals?



XCom -> RE: Reliever's Appearances way too high! (11/23/2006 4:53:22 PM)

Shaun, in my experience using fictional players the number of appearances seems reasonable but the overall number of innings for certain relievers is high. Often I have some relievers pitching in 50-70 games but throwing 150-180 innings all in relief. I've even had a few relievers win 20 games without making a start. I think this might be due to the pitcher hook lines in the XML where starters are being pulled very early in starts where they give up a couple of runs in the first inning. The reliever then comes in and typically will throw 5+ innings. My overall number of CGs is good, so there is not a general issue with pulling starters too early in games where they are pitching well. I'm working on tweaking the XML file now to try and fix this in my testing.

Anyway, it may be a separate issue than what sposfan brought up above...but just wanted to bring it to the discussion.



puresimmer -> RE: Reliever's Appearances way too high! (11/23/2006 5:32:37 PM)

Yea it looks like appearances is ok with fictional players (I just looked through a number of my test leagues).

So I'll have a look at real player usage.

On XCom's point, yea, I have seen the issue of them getting too many innings.

This is such a tough one to test/track down because there are multiple variables in play here, for example reliever usage can be greatly impacted by the composition of the AI's pitching staff, which is impacted by the draft and trade AI, which itslef is impacted by the endurance ratings assigned to players etc.

Then after I make a tweak I have to run 20+ year regression tests with real *and* fictional players, rinse, repeat. I sound a bit whiny don't I!

Shaun



XCom -> RE: Reliever's Appearances way too high! (11/23/2006 7:09:23 PM)

I totally understand, Shaun. You amaze me with how much you are able to get accomplished with so much other stuff going on around you. [:)]

I think that if we can find something to stop the games where the starter is pulled after the first inning only giving up 2 or 3 runs, then that could help with this problem. It's obvious that these pitchers are getting into games early and then racking up 5, 6, 7 innings at a time. I can understand starters being pulled after giving up say 5 or 6 runs in the first inning...but usually MLB managers will not make the change unless they have to so that they don't overwork their bullpen. BBPro used to have a concept called "toast points" in their pb.ini file to adjust when a pitcher should be pulled from a game. I don't think it worked very well in practice the way it was incorporated into their engine but I always thought the concept was pretty good.

I will be playing around with the hook lines in the XML to see if that makes any difference with this.



puresimmer -> RE: Reliever's Appearances way too high! (11/23/2006 8:55:45 PM)

Actually PureSim uses the exact same "Toast Points" concept. (I use to be a pb.ini geek with Front Page Sports Baseball -- consider it the inspiration for puresim.xml)

I think you are right about that early inning hook issue, that could be it.



XCom -> RE: Reliever's Appearances way too high! (11/23/2006 11:57:25 PM)

Cool. Are the toast points something that can be editable in the XML file? I don't see any lines in there for that. All that I've found that might be related is this line:

<PITCHER_HOOK hookVal="-1" hookValRelief="-2"/>

which I assume is a multiplier factored into the toast point calculations behind the scenes?

I haven't fully tested this line yet, but the concern I had was that if I adjusted the starter hook value so that pitchers were not pulled out of games too early when they were struggling that it would adversely effect pitchers being pulled out of games when they were pitching well...and thus mess with CGs.

The pb.ini file made BBPro infinitely playable, IMO, because you could always tweak the sim engine to generate the stats you wanted (with some work). I was thrilled to see something similar in Puresim and really appreciate you going to the effort to develop this area!

quote:

ORIGINAL: puresimmer

Actually PureSim uses the exact same "Toast Points" concept. (I use to be a pb.ini geek with Front Page Sports Baseball -- consider it the inspiration for puresim.xml)

I think you are right about that early inning hook issue, that could be it.




sposfan -> RE: Reliever's Appearances way too high! (11/24/2006 12:34:27 AM)

Shaun, thanks as always for your quick response. I am NOT using fictional players which would explain it. I appreciate you looking into this, but like I said, it's not a showstopper by any stretch of the imagination. The fact that this is the only stat that seemed to stick out is further proof how far puresim has come. I've played this version more than all the past puresims put together so you're definitely doing something right!




KG Erwin -> RE: Reliever's Appearances way too high! (11/24/2006 4:19:49 AM)

The issue of reliever appearances I believe is directly related to END ratings. I don't think we have enough CGs in historical leagues. It's a tricky thing, as we have to balance these out for the historical era being simmed.

For my 1946-60 Golden Age, during the reserve clause era, it was not uncommon for starters to complete at least half of their games.

For example, "Boo" Ferriss of the '46 Red Sox -- out of 35 GS, 26!! were CGs. SO, I don't think the "quick hook" is necessarily the main problem.

Tha main problem is giving END ratings for SPs that are too low. IMHO, this is the fix. Raise those END ratings for SP. The reason why the relievers are being brought in so soon is that the starters are wearing out too fast.

Whaddaya think, guys?



XCom -> RE: Reliever's Appearances way too high! (11/24/2006 5:09:56 AM)

Hey, KG. Thanks for the input and welcome to the discussion. I think you may very well be correct for replay leagues. It is possible that the algorithm used to convert Lahman data to player ratings may be a bit light on EN for older leagues and that would definately cause a lack of CGs for the era you like to play. In my experience for fictional leagues, I think that CGs work out quite well for the era I am trying to simulate which is the late 80s/early 90s before the big HR spike after the strike in 1994. I typically see my leader with 12-18 CGs and at least 10 pitchers with 10+ CGs each season...so I am happy with that.

As an example of the kind of results I'm describing for relievers...in my last test season I had this player:

P Lance Taylor 19-8, 2.74 ERA, 183.0 IP, 53 G, 0 GS
2nd on the team in IP in relief slot #1 with an EN rating of 53

Looking at his game log from this season he pitched 4+ innings in a game 24 times! And of those 24 games, the starting pitching gave up more than 3 runs only 7 times before they were taken out...typically in 0-2 innings worth of work. In those 24 games I would say realistically the starter should have been lifted when he was maybe 10% of the time. Most starters are superior pitchers to standard middle relievers. How many times do you see a pitcher give up 3 runs in the first inning and then pitch really well for the next 6 innings? It happens quite a bit and I don't think Puresim is giving those starters a chance to do that. Even after giving up 3 runs in 1 inning the starter will not be "tired" such that their performance will suffer if they stay in the game. Of course going too far here runs the risk of the AI leaving starters in way too long and unrealistically getting beat up if they "don't have it" that day. So, we just need to find the right balance.[sm=00000436.gif]



KG Erwin -> RE: Reliever's Appearances way too high! (11/24/2006 5:29:57 AM)

For my given historical league, and using real players, the ratio of CGs is off. So, I think my concern is valid.

However, the XML DOEs allow for situation-specific responses, so I believe that an all-inclusive XML, with conditional parameters, CAN be done.



XCom -> RE: Reliever's Appearances way too high! (11/24/2006 6:09:26 AM)

Hey, Shaun...are these lines able to be activated again in the XML:

<!-- Pitcher Usage -->
<!-- Not implemented in PureSim 2005, but will likely be revisted for 2006 -->
<!--
<PITCHER_USAGE RangeMin="0000" RangeMax="1871" usage_mult="0.4141" cg_mult="01.54"/>

etc...

-->

where the range values are year.

It seems they are commented out in the default XML file. Could these lines be used to tweak CGs for specific eras? Could be an answer to KG's issue.



puresimmer -> RE: Reliever's Appearances way too high! (11/27/2006 1:34:34 PM)

Guys, grab the latest 1.41 BETA patch, I have this one looking much better. Thanks for the posts!

Shaun



sposfan -> RE: Reliever's Appearances way too high! (11/27/2006 2:13:14 PM)

Already done, thanks for the quick turnaround! And as an added bonus, I love the way it now treats aging vets, not making them end their careers languishing in the minors!



XCom -> RE: Reliever's Appearances way too high! (11/27/2006 4:35:02 PM)

Shaun, thanks for the support on this issue. Right out of the box I was not able to see too much difference in my fictional test league. Appearance numbers were fine but I was still seeing some relievers pitching an inordinate amount of innings. In my first test season with 1.41 I had a reliever go 22-9 in 63 appearances and 175 IP. But, since then I've adjusted the hook line in the XML and have gotten much better results. In my last test season the top reliever had 138 IP and the the leader in appearances had 69. I need to run several more seasons to make sure this trend holds, though. I'll post back here once I get a few more seasons under my belt.



sposfan -> RE: Reliever's Appearances way too high! (11/27/2006 5:21:10 PM)

XCom,

what settings are you using for hook values?



XCom -> RE: Reliever's Appearances way too high! (11/27/2006 6:48:52 PM)

Currently, I'm using

<PITCHER_HOOK hookVal="-7" hookValRelief="2"/>

I need to run some more tests tonight to make sure what I am seeing is a real trend and not an anomoly.

quote:

ORIGINAL: sposfan

XCom,

what settings are you using for hook values?




sposfan -> RE: Reliever's Appearances way too high! (11/28/2006 12:41:10 AM)

Thanks XCom, looking forward to your findings! Don't forget to make sure you have the updated 1.41 in case, like myself, you downloaded it right away!



XCom -> RE: Reliever's Appearances way too high! (11/28/2006 7:26:42 AM)

Thanks for the reminder on re-downloading the patch, sposfan. I did have the wrong version and have since gotten the re-release. My last test season really highlights how well this is working out now. My leader in appearances had 56 (maybe a bit low but I'm okay with it) and my top reliever had 131.1 IP. Not bad! [8D]

I'm going to run some more test seasons and see how reproducible this is, but I like the results so far.

quote:

ORIGINAL: sposfan

Thanks XCom, looking forward to your findings! Don't forget to make sure you have the updated 1.41 in case, like myself, you downloaded it right away!




sposfan -> RE: Reliever's Appearances way too high! (11/28/2006 2:42:18 PM)

No sweat! Did you re-adjust your hook values, since each patch will write over your modded XML with the vanilla one.



XCom -> RE: Reliever's Appearances way too high! (11/28/2006 4:14:35 PM)

Yes, I keep a copy of my edited XML file in another folder so it doesn't get overwritten. I will do some more testing tonight and let you know how it goes.

quote:

ORIGINAL: sposfan

No sweat! Did you re-adjust your hook values, since each patch will write over your modded XML with the vanilla one.




bobsayah -> RE: Reliever's Appearances way too high! (11/28/2006 9:27:35 PM)

OK, I know I'm late to the party here, but...

While there may be an issue with the code which controls when the starter gets pulled (and which I guess is fixed now), I think another contributer to this is that the AI has a tendency to use the #1 reliever a lot in cases where the starter gets pulled early. In real life, during the course of the regular season, when the starter gets pulled early, you will typically see the guys at the bottom of the bullpen get called in (except perhaps for playoffs and "must win" late season games). The #1 guy typically gets saved for another day, when the game is close in the late innings. I think this is a big contributer to the large number of innings that the #1 relievers are getting.



XCom -> RE: Reliever's Appearances way too high! (11/28/2006 10:27:21 PM)

I think that's a great point, bob. You are correct that Puresim will often bring in the #1 reliever in cases where the starter is knocked out early. When I look at the reliever priority list I tend to think that my #1 reliever (the best on the list) should be the setup guy in associations using the modern closer option.

Not to keep bringing up BBPro (it really did have a very nice setup and good solutions to a lot of issues), but it used a system where you could designate bullpen roles to your relievers. Thus, you could place that #1 reliever into the setup role and he would be used primarily there. Now, of course as was the case with a very many of BBPro's features, the concept was good but the implementation was not. The computer manager never really got it right based on your bullpen designations. But, maybe something like this could help tighten up bullpen usage in Puresim.



sposfan -> RE: Reliever's Appearances way too high! (11/29/2006 6:57:39 AM)

XCom, you can, in a fashion, do what you are talking about through the player usage card. I'm doing this from memory so bear with me, but I believe on the left side, about 1/2 way down, there is a pulldown menu where you can set a relievers role. The choices are (I think) use primarily as a setup, mopuo or long reliever. Granted this would only work for human controlled teams, but maybe Shaun can have the CM designate it's pitchers as well.



XCom -> RE: Reliever's Appearances way too high! (11/29/2006 7:45:26 AM)

Ahh, very cool. Thanks for the heads up, sposfan. There are so many facets to this game, I think I'll still be discovering little nuggets years from now. [:D]

Didn't get a chance to run any test seasons tonight, but hoepfully will run some tomorrow night and I'll keep you posted on my results.



XCom -> RE: Reliever's Appearances way too high! (11/30/2006 7:16:08 PM)

Ran a couple more seasons last night and this trend on lower reliever usage is holding up. In my three seasons with the new settings the highest usage for a reliever was 141.1 IP. Still a bit low on the number of appearances, I think max was around 65. My complete games seemed to have gone up just a touch...but overall good results with the new patch and the XML tweak.

quote:

ORIGINAL: XCom
Didn't get a chance to run any test seasons tonight, but hoepfully will run some tomorrow night and I'll keep you posted on my results.




sposfan -> RE: Reliever's Appearances way too high! (11/30/2006 11:29:42 PM)

Is this with your modified hook value or with the as-shipped XML?



XCom -> RE: Reliever's Appearances way too high! (11/30/2006 11:36:18 PM)

Modified hook values. I may increase the starter hook a little to see if that brings CGs back down to where they were. Seems saves are down a touch too, so I don't want the hook modifer to be so low that we are seeing starters finish too many games at the expense of the saves numbers. But, overall these numbers are off by 10% or less so I'm nit-picking. [;)]

quote:

ORIGINAL: sposfan

Is this with your modified hook value or with the as-shipped XML?




sposfan -> RE: Reliever's Appearances way too high! (12/1/2006 2:30:37 AM)

keep us posted!



XCom -> RE: Reliever's Appearances way too high! (12/2/2006 8:58:42 AM)

Hmmm...may have found something that was artificially altering the trends I thought I was seeing.

After simming another couple of seasons with increased hook values for starters I did not see CGs being brought back down to where they were before...in fact overall CGs continued to slowly increase. I thought that was strange, so I decided to try my tweaked XML on a new fictional association to see if the trend was being caused by something other than the XML lines. Lo and behold in the first season with the new association I had CGs way down and the reliever IP problem was back! I had one reliever pitch 67 games and 200+ IP. Back to square one...well, maybe.

I thought maybe it was the ratings that were causing this bullpen usage effect. When I looked at the pitchers in the new league and sorted them by EN I found that just about every pitcher with EN over 90 was less than 25 years old. And there were only a handful of pitchers in their "prime" years (27-31) with an EN rating in the 80s. So, then I went back to my previous test association with 10 years of history. Did the same pitcher sort by EN and now about 60% of the pitchers with EN over 90 are age 25-30. Those are the pitchers who are getting the increased number of CGs as they mature and become regular starters. Now, I think this directly affects the reliever usage. More starters with higher EN values means the starters pitch more innings and overall reliever usage goes down.

Is this what was causing the reliever results I thought were so good earlier this week? Not fuilly sure yet. I'm going to setup some auto play overnight tonight to sim as much history as I can with a new association to see how the stats change over time and see if I repeat the increased CGs and decreased reliever usage after season 4 or so.

If I do see the same trend then I figure the next thing to try is to increase the weights slightly for EN in the player generation lines for pitchers age 25-30 and decrease slightly for pitchers age 17-24. Hopefully smoothing out the distribution of EN ratings and maybe reducing the median value a touch. Anyway, let's see what happens with the test.



Page: [1] 2   next >   >>

Valid CSS!




Forum Software © ASPPlayground.NET Advanced Edition 2.4.5 ANSI
0.609375