How good is the AI at drafting?

[nino33]

Hello everyone


Along with wondering about how good the AI is at resigning it's RFAs http://www.ehmtheblueline.com/forums/vi ... 86#p223686 I'm wondering how good is the AI at drafting in the NHL Draft...what do you think? If you think it could use some improvement, how?


Here's the start of a discussion about the subject from today on HFBoards...

• Lebowski said But you don't need to exploit the game to build super teams... That's the thing.
  You end up recognizing some patterns in prospect development that makes it easy to build crazy strong prospect pools.

• and I replied One of the testing tasks I've wanted to get to lately is looking at the NHL Drafts and specifically at the AI abilities in the NHL Draft - but I just haven't got to it (it's a lot of time and effort, and my motivation hasn't been high for such things lately + I've been enjoying doing other things in my life too)
  
  I had started on the task, and what I thought I found was it seemed at first glance that to many of the Very Good/Top players were to low in the Draft - essentially the top 10 picks were often filled with guys who should be much much lower
  
  It seemed to me this allowed the human player to have far more success than the AI; it wasn't just that the AI would select players with 2-3 very weak Attributes (that a human player would never select) and thus end up with a dud, but that the AI wouldn't select the "best player available" by just looking at the visible Attributes
  
  
  Personally, I wish there was a setting where you could check it off at startup and as a result the AI would take the best PA player available 70-90 percent of the time OR the standards humans use in online games could be coded in (the order of selection would be the handful of "priority Attributes" (like Work Rate, Determination, Anticipation and minimum numbers in skating) + the highest total of all visible Attributes would get chosen
  
  This way, if users wanted it to be easier they could leave it "as is" but otherwise they could check off the setting that makes the AI a good/very good drafter

If you think it'd be nice if the AI was a lot better at drafting, can you give some criteria that you would suggest...what are the "most important Attributes" to look at? Is there a minimum (low) number in any/all Attributes that "automatically should" makes a player not an early round selection? Maybe this "better AI drafting" should be just for the first round?...or the first 2-3 rounds?


Thanks for your time everyone

[Primis]

I had started on the task, and what I thought I found was it seemed at first glance that to many of the Very Good/Top players were to low in the Draft - essentially the top 10 picks were often filled with guys who should be much much lower

It seemed to me this allowed the human player to have far more success than the AI; it wasn't just that the AI would select players with 2-3 very weak Attributes (that a human player would never select) and thus end up with a dud, but that the AI wouldn't select the "best player available" by just looking at the visible Attributes

I can chime in here just a bit.

AFAIK, this is by design. Riz designed the CSB in the original freeware EHM's to be inexact, and for some guys to be in every Top 10 that are guaranteed to be busts (you'd always get Enforcers and Character players in the Top 10, and those two roles could never ever possibly develop to their potential in the freeware, for example). In real life, if you look at the Top 10 of every draft there are usually some total busts in there, and these are kids that were in the CSB/ISS Top 10 all year long before the draft. The AI is taking these because they're supposed/designed to. Because NHL teams do exactly that, and they pick duds. IIRC, the original CSB rankings in the freeware were done by summing up all of a players' attributes, including POT, CON, and Greed (which were hidden), and then dividing by the number of attributes or something. It was done deliberately to confuse the AI a bit, and also to put in some landmines to players if they did not properly scout the draft (because it wasn't just Enforcers and Character players that were duds).

That obviously is not the perfect method. But AFAIK, Riz has used a similar approach with the drafts on up through EHM 1 with some tweaks. Essentially, he's tried to replicate how real life goes, rather than make the AI the absolute smartest. The NYR drafted Dylan McIlrath in the Top 10 even though *at the time* most knew it was a really bad plan. And those people were very, very right. Or to stick with the NYR, Hugh Jessiman at #12 Overall, as another example (even though it's just outside the Top 10).

Then factor in the scouting and evaluating abilities of AI staff, and you have another layer of variables still... basically another opportunity of where if you're rolling dice a Critical Fail can occur.

[Shindigs]

There is a very simple way to improve the AIs drafting if you want to. The AI "cheats" off your shortlist. So just shortlist the best players in a draft and it will increase the likelihood the AI picking them. This is also true in non-draft leagues. If you play in Swe-2 and start shortlisting belarusian players, which is a good thing to do since they have high potential pa and low rep. The AI will start looking into signing belarusian players too. Whereas if you don't scout/shortlist belarusian players. You will never see one in Swe-2 ever.

But if you wanted the AI to be better at drafting by default it would be quite easy too.
Make the AI favor these traits: High consistency, High temperament, Strong Character, Competitive, Relaxed.
...and ignore picking players with: low consistency, low temperament, weak character, unambitious, slack, <12 Determination, <7 Work Rate, <7 Teamwork.

What I personally do in drafts is I filter out all sub 12,7,7 mentals players so I won't ever pick them. I also put "low con" in the note of any player with low consistency mentions at all in any of their scout reports, since consistency sadly can't grow:/

When the AI proceeds to top pick players with horrible mentals and low consistency I just open the top 3-5 draft picks in the db editor and change their low con to good con, and up their mentals to acceptable values. So the AI's top picks actually become top picks. I do the same with the AIs "draft steals" once they get to the NHL level and look like they'll play. If they have nonsense holes in their mentals I just up it to what a player of that PA "should" have. And the AI is pretty decent at finding the draft steals, they often will snipe potential steals I have lined up for the 120-210 range in the 2nd/3rd round. Which is a reasonable place to pick them, I just draft very greedy so I try to save them for later.

The change to newgen and regen mentals and mentals growth in 1.4 has made drafts much better than they were though. In 1.3 there were years where every single player had one of the things that made them undraftable. Those years were so awful. I ended up just getting horrible players with High scouted PA to trade them for picks in following years. Low consistency also seems to be less common, but I could just be having good variance in this save.

And although not to do with the draft per-se I think the game would be better off if the Enforcer role no longer was found on newgens. Since that role is all but dead in real hockey.

[Shindigs]

Another update on the AIs drafting skills. I've started to more aggressively help the AI get back in the game by making it so the bottom teams' 1st draft picks are always good. Also we're running out of capable goalies fast in my save so I'm needing to mess with those as well for the AI's sake. And the league as a whole. Even with me buffing the non-drafted goalie prospects it just wasn't enough. Currently the guy rated as the 4th best goalie in the NHL is having a whale of a time trying to beat .900 on a consistent basis. The perfect storm required to create a .910+ goalie is really, really rare. Just PA is far from enough, the attribute requirements are so strict.

Both of the last two drafts the AI team with the #1 pick has ignored ISS rankings and drafted someone in the 5th to 10th range 1st overall. The first one was a dman with 124 PA. So I manually changed him to 160 so the poor AI can get back on track. The year after a goalie that had a grand total of 0 of the prerequisite mentals to be a top goalie and a PA of 156 got picked 1st overall. So I gave him +2 to all mentals and set his PA to 180. Only after doing so I realized that it was the Rangers who picked him, guess Canadian Henrik Lundqvist will carry them on his back now haha.

Edit: The rangers got 1st overall pick again, went for the 9th overall on ISS rankings. His scout report in-game says "more talented Pierre-Luc Dubois", which is essentially ~180 PA. But he's a bust so he's only actually got 128 PA. Man I'm having to help the rangers so much. They are the masters of picking draft busts. Now I've given them a 180 PA goalie and a 176 PA 2way center, if they still can't stop finishing dead last I give up on even trying to help them.

[qa2wer]

I also put "low con" in the note of any player with low consistency mentions at all in any of their scout reports, since consistency sadly can't grow:/

Are you 100% sure about the consistency not growing? I tend to remember seeing some regens developing couple of points on consistency, but this may have been in the ehm07 rather than EHM1. Anyways I think it's a bit odd that it doesn't grow by age atleast a little bit, as the older guys are commonly more consistent than the young guns in real-life.

[Shindigs]

I also put "low con" in the note of any player with low consistency mentions at all in any of their scout reports, since consistency sadly can't grow:/

Are you 100% sure about the consistency not growing? I tend to remember seeing some regens developing couple of points on consistency, but this may have been in the ehm07 rather than EHM1. Anyways I think it's a bit odd that it doesn't grow by age atleast a little bit, as the older guys are commonly more consistent than the young guns in real-life.

I'm going by the very extensive testing nino33 did on the matter. Also I have never seen it grow on any player. Important matches grows like wildfire if you make the playoffs enough. But Consistency just is what it is. A low consistency player can still be good. But why draft one when you could have one that's consistent?

And I completely agree on that it should grow. In real life consistency generally isn't something you're born with, it's the result of having a lot of pre-requisites that make you consistent. Some players just cannot become consistent no matter how long they play, true. But generally experience is a large part of what causes real life consistency, so most players should become more consistent as they become more experienced.

The actual consistency of your player in the game comes from so much more than just consistency though. Just because a player is consistent, it doesn't mean he's consistently good. There is such a thing as being consistently bad (just look at the Coyotes)

[Primis]

And I completely agree on that it should grow. In real life consistency generally isn't something you're born with, it's the result of having a lot of pre-requisites that make you consistent. Some players just cannot become consistent no matter how long they play, true. But generally experience is a large part of what causes real life consistency, so most players should become more consistent as they become more experienced.

Just looking at it from a gameplay perspective:

I don't think game-players want the flip-side to that, though, a guy's Consistency falling. In the freeware version, a guy's CON could plummet (halve or more) after a trade. It was very annoying, because here you are trying to help your team acquire a good player, and then his CON dropped and he was no longer useful. Many freeware leagues simply started posting player CONS pre-trade, and then manually fixing/returning them back to that after processing the trade in-game.

So I don't know that you want guys growing in Consistency much at all, because you then need the opposite end of that or else you end up with a game full of players wth good consistency.

[Shindigs]

And I completely agree on that it should grow. In real life consistency generally isn't something you're born with, it's the result of having a lot of pre-requisites that make you consistent. Some players just cannot become consistent no matter how long they play, true. But generally experience is a large part of what causes real life consistency, so most players should become more consistent as they become more experienced.

Just looking at it from a gameplay perspective:

I don't think game-players want the flip-side to that, though, a guy's Consistency falling. In the freeware version, a guy's CON could plummet (halve or more) after a trade. It was very annoying, because here you are trying to help your team acquire a good player, and then his CON dropped and he was no longer useful. Many freeware leagues simply started posting player CONS pre-trade, and then manually fixing/returning them back to that after processing the trade in-game.

So I don't know that you want guys growing in Consistency much at all, because you then need the opposite end of that or else you end up with a game full of players wth good consistency.

To be fair the issue there sounds more like the rate at which it could grow/decline was too high. The issue isn't that it can change, but rather the way in which it did wasn't balanced. I guess Riz opted for removing it rather than balancing it. Which I guess is understandable with his time-constraints.

The fact that it can't ever drop is somewhat of an issue as well, like nino33 touched upon (somewhere) goalies shouldn't be equally consistent from season to season, since they really aren't in real hockey. When you get a good, consistent player in EHM he's good and consistent forever. Especially now that losing their legs seems to have much less negative impact than it used to (in my experience). If you draft a consistent goalie at 17-18. He will still be equally consistent at 37 no matter what. That's way too powerful and a large reason as to why a few years into a save the draft is near pointless, because the conversion rate from prospect to consistent starter is way too high. At some point you'll have more top 6 forwards and top 4 defencemen than you can throw a stick at.

You still won't have any goalies though, cause they don't regen/newgen with the right attribute spread. 10 years into my current save there is a grand total of 0 newgen/regen starters in the NHL, and only 8 goalies who were undrafted at the start. For all but one of those I had set their core attributes in the db to give them at least a shot at being a real goalie, assuming they highrolled their PA. Even with me doing that there is a notable decline in goalie quality from the start of my game until now. 15 of the teams use a backup/third stringer or worse as their starter. And since the draft isn't refilling, the issue gets worse by the year. We are coming up on the last of the current starters retiring now, and there is absolutely no sign that anyone will be able to fill their shoes. To the point that I'm needing to manually go through the goalies in the drafts and fix their core attributes to be of a proper spread, so maybe the AI can get real goalies again. Me and Calgary both hit the jackpot in the first draft so we're fine. But not so much for the other 28 teams. I really shouldn't need to cheat this hard on behalf of the AI to get the goalie quality to an acceptable level. But I do. Meanwhile Vasilevskiy is still in the KHL and Tampa is using Hellebuyck as their starter, cause that seems reasonable.