The new user interface is in preview!

Want to check it out? Click here! (If you don't like it, you can still switch back)

League Forums

Main - General MFN Discussion

Re: AI Offseason Overhaul Thread

By bgedgerly
9/11/2019 3:25 pm
I am making a new thread since the original comment was at the tail end of a post that had nothing to do with this issue. JDB had commented that he would be open to ideas on how best to fix the problems surrounding AI-controlled teams leading up to and during the offseason, specifically with regard to free agency, salary cap management, and the draft. We all know that the CPU currently destroys teams without human owners, whether its by letting superstars walk, handing out outrageous contracts to players and coaches, or drafting a bunch of worthless players with high picks. This issue has already massacred pretty much every MFN league already, and is becoming a problem in custom leagues, even ones tightly managed. I have some ideas for fixing this problem that JDB would of course have to work into the code, but I wanted to give everyone a chance to put in their opinions as well. I'd like everyone to stay on topic as much as possible, and let's be polite. There are no stupid questions, and as long as everyone is trying to work toward fixing the issue at hand, there are no stupid suggestions. I look forward to a healthy discussion (as I'm sure JDB is), and below you will find my outline for fixing the issues surrounding AI-controlled teams.

--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------

As far as the AI goes involving team management, I don't know how to fix that. I would say that if the team has the cap room/cash flow to re-sign all of the players/coaches at season's end based on their minimum salary demands then the AI will do so.

IE, if John Doe won't negotiate for a contract with less than a 5 million signing bonus, the computer re-signs him at the lowest contract price with a 5 million dollar signing bonus.

And if the team DOESN'T have the cap room to re-sign everyone, the AI would prioritize based on highest rated players (per default ratings) to lowest. So if the team in question had a 93 rated player and a 72 rated player, but only the money to re-sign the 93, that player would get re-signed and the 72 would be left to free agency.

I don't know what kind of coding cost it would require, or if its possible, to have the AI make decisions based on age (Don't re-sign Year 11 players to 6 year contracts), but I think an age based system on top of a salary cap triage system would be ideal, because the dead money caused by early retirements is a big reason behind AI team cap issues, although admittedly, a lot of actual players deal with this issue as well.

Coach hiring would just be whoever the AI could afford to hire to stay under cash flow cap, maybe based on average salaries league wide, or just sign whoever is available before training camp. I honestly don't have much of an opinion on coaches, other than to have the AI not hire a QB Coach for 99% of cash flow.

BUT, as far as the draft goes, my idea would be that the AI drafts the best player available based on position of need (PON). It already handles PON pretty well, it just really craps its pants when it comes to actual talent. So if a team has 13 OL and 2 WR, the AI should be designed to draft the best WR before it drafts the best OL. And so on. And, just to make it easier (I'm guessing), just write drafting punters and kickers out of the AI code altogether. If it's an AI controlled team, the K and P positions can be filled with the 34545432 free agents. That way, if a real person comes to take over a team that the AI drafted for a few years, there's a better chance of that player contending if the AI was at least trying to take players based on value. Volatility really shouldn't factor in, just to make it simpler. If they bust, they bust, but there's no chance of success if the AI drafts 3 RBs rated 37 overall with 40 speed.

So, to wrap up, this is my ideal AI position:

1. Re-sign players to contracts as small as possible based on player demands.
2. Prioritize re-signing players with higher default rankings.
3. Prioritize re-signing younger players over older players.
4. Be as frugal as possible with both player and coach contracts. Saving money should be the biggest priority for AI controlled teams.
5. No drafting K or P.
6. Prioritize drafting based on best player available (according to default ratings, disregarding volatility).
7. Secondary drafting priority is PON.

If any of this isn't possible, or if anyone thinks any of this is stupid, feel free to let me know. But if these concepts are reasonable then I think MFN in general will be a much healthier community with them in place.

Thanks again for everything JDB.
Last edited at 9/11/2019 3:29 pm

Re: AI Offseason Overhaul Thread

By TarquinTheDark
9/11/2019 3:46 pm
It would help with the draft issue if default ratings were reworked. Nobody wants a team loaded with default 75+ rated RBs, WRs, and DBs ... and average speed of 60.

Re: AI Offseason Overhaul Thread

By bgedgerly
9/11/2019 3:55 pm
TarquinTheDark wrote:
It would help with the draft issue if default ratings were reworked. Nobody wants a team loaded with default 75+ rated RBs, WRs, and DBs ... and average speed of 60.


I completely agree. I think that reworking the default rankings to be more speed-oriented (like most of our adjusted weights) would be ideal. And if while coding the new AI changes its an easy fix to adjust the weights, I think that the trigger should be pulled on that. If not, I'd rather get the current system fixed first and then have the weights adjusted later.

And if my responses sound vague, its because if coding was a part of the Wonderlic I'd be scoring in the Morris Claiborne percentile.
Last edited at 9/11/2019 3:55 pm

Re: AI Offseason Overhaul Thread

By punisher
9/11/2019 3:55 pm
maybe this would work for CPU teams if they signed players like if they were owned by a person

here = https://atrealnfl.myfootballnow.com/community/5/5651?page=1#34039

load up what a person would do for draft for CPU teams

here = https://atrealnfl.myfootballnow.com/community/5/5427?page=1#32741

Re: AI Offseason Overhaul Thread

By bgedgerly
9/11/2019 3:59 pm
punisher wrote:
maybe this would work for CPU teams if they signed players like if they were owned by a person

here = https://atrealnfl.myfootballnow.com/community/5/5651?page=1#34039

load up what a person would do for draft for CPU teams

here = https://atrealnfl.myfootballnow.com/community/5/5427?page=1#32741


As someone in the first thread mentioned, I'm not crazy about programming anything to mimic certain players, but definitely adding in 'player-esque' biases towards positions (DB over DL, RB/QB over OL) and biases towards skills (Speed over Intelligence, Throwing Accuracy over Ball Carrying, etc.) in the default ratings rework would be wonderful.

Re: AI Offseason Overhaul Thread

By TarquinTheDark
9/11/2019 4:05 pm
As a former coder, I believe reworking weights would be much lower-hanging fruit (less work for the payoff) than balancing primary vs. secondary drafting priorities the way a human can.
Last edited at 9/11/2019 4:06 pm

Re: AI Offseason Overhaul Thread

By punisher
9/11/2019 4:09 pm
bgedgerly wrote:

As someone in the first thread mentioned, I'm not crazy about programming anything to mimic certain players, but definitely adding in 'player-esque' biases towards positions (DB over DL, RB/QB over OL) and biases towards skills (Speed over Intelligence, Throwing Accuracy over Ball Carrying, etc.) in the default ratings rework would be wonderful.


dont think anyone is crazy about doing it either but i would think it is unavoided in that this whole thread i thought is for any idea that improves CPU from being Hot Garbage (signing players that are going to retire to big contracts , signing players that probably too slow for their position , etc. ) to being at least respectable (basically where people would not have to have other accounts for their leagues so they can take over teams so CPU wont screw team up where no one will take it over , CPU wont sign players that are going to retire or at least they wont give them a contract that messes up their salary cap big time , etc. ) so you basically have to start somewhere (maybe things that may not work , maybe things are silly , etc. ) to where you get to place where the idea or ideas actually will work , improve the CPU to the point it is like a difference between day and night , etc.

Re: AI Offseason Overhaul Thread

By TarquinTheDark
9/11/2019 4:22 pm
A human can also work out issues such as:

I really need to replace my QB before he retires, but if I don't get a DB now, there won't be any left. There are three teams in my league with more than two QBs, so I should be able to trade for one within the next two seasons.

or:

This team sucks, so I'm just going to draft defense, sign a whole bunch of OLs and FBs from the FA market, and try to trade up.

The AI can't trade, so it shouldn't exactly be trying to mimic human drafting behavior.

Re: AI Offseason Overhaul Thread

By bgedgerly
9/11/2019 4:24 pm
punisher wrote:
bgedgerly wrote:

As someone in the first thread mentioned, I'm not crazy about programming anything to mimic certain players, but definitely adding in 'player-esque' biases towards positions (DB over DL, RB/QB over OL) and biases towards skills (Speed over Intelligence, Throwing Accuracy over Ball Carrying, etc.) in the default ratings rework would be wonderful.


dont think anyone is crazy about doing it either but i would think it is unavoided in that this whole thread i thought is for any idea that improves CPU from being Hot Garbage (signing players that are going to retire to big contracts , signing players that probably too slow for their position , etc. ) to being at least respectable (basically where people would not have to have other accounts for their leagues so they can take over teams so CPU wont screw team up where no one will take it over , CPU wont sign players that are going to retire or at least they wont give them a contract that messes up their salary cap big time , etc. ) so you basically have to start somewhere (maybe things that may not work , maybe things are silly , etc. ) to where you get to place where the idea or ideas actually will work , improve the CPU to the point it is like a difference between day and night , etc.


Coding limitations are a big bugaboo with this issue, and while I would love for the AI to have some 'personality', I think that it will probably be an easier (and therefore quicker) fix to just adjust the offseason move tendencies away from the current disaster to something more manageable.

Remember, the point of this fix won't be to make the AI-controlled teams title contenders, its to keep the open teams fixable by new owners and therefore attractive rebuild projects instead of just being a toxic pile of salary cap penalties and blown picks that no one wants anything to do with.

I don't mean to constantly disagree with you, punisher. I think that you have a good idea with regard to AI variation, but I don't think that the technology is currently available for that level of individual automation...or at least, not here. I do think that each code should bring new updates to the AI positional weights just to keep the human-controlled teams honest.
Last edited at 9/11/2019 4:27 pm

Re: AI Offseason Overhaul Thread

By punisher
9/11/2019 4:41 pm
bgedgerly wrote:

Coding limitations are a big bugaboo with this issue, and while I would love for the AI to have some 'personality', I think that it will probably be an easier (and therefore quicker) fix to just adjust the offseason move tendencies away from the current disaster to something more manageable.


well in that case how about limiting CPU to only 46 players where they have enough players for each position and cover the whole depth chart yet maybe if they are limited to that number they cant mess up the salary cap unless of course the CPU can and will mess up the salary cap with 46 players on it.

bgedgerly wrote:
Remember, the point of this fix won't be to make the AI-controlled teams title contenders, its to keep the open teams fixable by new owners and therefore attractive rebuild projects instead of just being a toxic pile of salary cap penalties and blown picks that no one wants anything to do with.


wasnt expecting that whatever is suggested that the CPU would be a title contender because to be quite frank if the CPU is going to be a title contender or be a team that be handling out whippings to people i would think people would turn away from game and leave this game to the point no one is playing it and JDB pretty much can turn off the lights because no one is here.
just wanted the CPU to not do boneheaded things like for example if i give 1 year contract to player that has 30 percent to retire what gets my blood boiling is the CPU will give that same player a 6 year contract and he may play 1 year or 2 but once he does retire the CPU will be on the hook for the rest of the contract.

bgedgerly wrote:
I don't mean to constantly disagree with you, punisher. I think that you have a good idea with regard to AI variation, but I don't think that the technology is currently available for that level of individual automation...or at least, not here. I do think that each code should bring new updates to the AI positional weights just to keep the human-controlled teams honest.


it is ok if you disagree with me i actually encourage it as long as people can disagree without the petty remarks , petty insults , etc.
So with that said i dont take offense to you disagreeing with me and keeping it professional about it