Revamping Draft Class Attributes (250 Players)

[AdamJones113]

Goals:

• To set up draft classes as a realistic continuation of previous classes
• To revamp the scouting and draft mechanism with a higher focus on stats and special skill as opposed to overall/potential

Index:
1. The Math
2. Potential
3. Revamping the Scouting Process
4. Scouting
5. Mock Draft


Part 1: The Math

The way it stands now, draft prospects are assigned attributes and potential based on real-life scouting (for players in the first several years of Be A GM) and then randomized attributes as the game sees fit (for fictional players in later years of Be A GM). Future draft classes have good players, bad players, great players, and terrible players, with the overall goal of maintaining the skill balance in the NHL. [Though from a personal perspective, the overall skill/potential of prospects increases as you get further into the mode.]

Four years ago, who would have predicted that there would be a guy like Conor McDavid? Four years ago, the CHL just got its second "Exceptional Status" player. It's incredibly hard to predict the future, so that's why the goal of simply maintaining the skill balance is smart and generally attainable. But the game needs to take away some randomness for that to work. If first rounders are generally the most successful, then your later top prospects need to be in that round. There's no way to recreate a Pavel Datsyuk or a Henrik Lundqvist.

Let's try to change that.

Let's assume that there are 250 prospects in the pool per year (30 picks per round * 7 = 210, +/- 40 players undrafted), a nice round number. We'll use the hypothetical 2025 draft as an example. Each skater has 25 individual attributes that have to be assigned a rating. Let's use passing as an example.

/Code that assigns a random number between 36 and 99
/Code adds one random ++1 (i.e. 1, 2, 3 etc) multiple up to 7 and subtracts --2 multiple (i.e. 2, 4, 6 etc) up to 14.
= Final rating for that attribute.

Ok, let's explain that mumbo jumbo. Obviously the code will shy away, as it were, from getting too close to 36 or 99 (the current code works fine for example). But the modifiers are designed to add randomness while ensuring that most players are lowly rated but there remains the possibility of a Conor McDavid. Your once-in-a-generation player will be the guy that gets the lucky dice roll and hits 7 added points and only loses 2.

Think of it in terms of "[first] round talent." So you can think of someone who winds up highly rated in passing as a first round talent in passing, but he'll probably have some problems elsewhere. Since there are 25 attributes to go through in the this manner, it's very rare that you'll see a superstar player right out of the draft. However, 7th-round flairs are more possible, since there's the element of randomness that allows them to be skilled in several areas.

Please note that I am simply trying to build on the current coding. Were this method to be the sole method, round would hardly matter (though the math is not over yet). EA's current proportions for draft placement are a crucial starting point.



Check back later for part 2, "Potential"

[BigDisk]

Re: Revamping Draft Class Attributes (250 Players)

Ran your algorithm. It's not pretty.



That's the distribution (occurence count) of 10,000 attributes.

You don't want to use a simple random number. Because it will definitely result in a uniform distribution. You always want this sort of distribution to be "normalized", so it looks like a Bell curve (https://en.wikipedia.org/wiki/Normal_distribution).

Here's a link your algorithm coded in javascript if you want to have some fun : http://jsfiddle.net/WbHsX/156/

The rnd() function is your randomly distributed attribute set (black series).
The rnd2() function is a pseudo-normally distributed set. It does look a lot better IMO (blue series). Not sure about the standard deviation though, too lazy to calcuculated but it looks like 80% of the results are between 60 and 75.

Edit : Not sure how familiar you are with programming, so I commented the jsFiddle if you want to change it to suit your needs.

[AdamJones113]

Re: Revamping Draft Class Attributes (250 Players)

Ran your algorithm. It's not pretty.



That's the distribution (occurence count) of 10,000 attributes.

You don't want to use a simple random number. Because it will definitely result in a uniform distribution. You always want this sort of distribution to be "normalized", so it looks like a Bell curve (https://en.wikipedia.org/wiki/Normal_distribution).

Here's a link your algorithm coded in javascript if you want to have some fun : http://jsfiddle.net/WbHsX/156/

The rnd() function is your randomly distributed attribute set (black series).
The rnd2() function is a pseudo-normally distributed set. It does look a lot better IMO (blue series). Not sure about the standard deviation though, too lazy to calcuculated but it looks like 80% of the results are between 60 and 75.

Edit : Not sure how familiar you are with programming, so I commented the jsFiddle if you want to change it to suit your needs.

Thank you for showing the people my behind the scenes work . Also, don't run 10,000 here. That's 40 draft classes. Try running several at 250—generally winds up a much more reasonable scale. At most, run 6,000 (24 draft classes, 25 seasons of GM Mode - known prospects). That works out to the essentially the same ratio (I am still tuning things), but...

I'm not done yet! That's not the only work there is to it, of course. Once I explain the system, you'll see that winding up with 99 overall player is harder than you think. The "Potential" section is going to revolutionize how you think about the scouting system in EA NHL.

Thanks for running it, though.

[BigDisk]

Re: Revamping Draft Class Attributes (250 Players)

It's stats mate. You want to test an algorithm with millions of tries to see distribution, ie., what are the odds of every attribute value being randomized.

So, basically, your random algorithm +1..7 -2..14 gives a flat distribution; as much chance to hit 99 than 70 than 66.

That's not how it works in real life. Most values are within 1-2 standard deviation of the average, pretty much all the time.

That's where a Gaussian blur helps to simulate reality, because it gives a bell curve type of distribution where 80% of the values are within a 20% range [-10%..average..+10%].

I'm just dumbing things down but I must admit that while your singles attributes are uniformly distributed, the resulting overall would be normally distributed if it's a simple average, but played stats would be all over the place.

I love algorithms, that's why I put my 2 cents here. I'll leave you bring forth your whole concept for the game now.

Peace

[AdamJones113]

Re: Revamping Draft Class Attributes (250 Players)

It looks like a lot, but I think there's some good stuff and good food for thought in the below section. I'd advise you to read closely—it's a good thing to do to pass the work day, lol.

Part 2: Potential

The potential rating, in two words, is completely unrealistic in EA NHL. To be more specific, telling users that there is a 'high' probability, or a 'low' probability, that a player will reach his potential is a major strike against reality. If the Trail Blazers knew that Sam Cassel had a 'low' chance to reach his potential, they probably would have gone for Michael Jordan. This designation I suggest, be, simply, taken out completely. What follows, however, is not as simple.

You see the even numbers above. BigDisk has thoughtfully pointed out that the overall of the NHL is rather like a curve, starting low, going high, and returning south once more. The numbers my equation spit out are very even. Therefore I reveal something that will lower Progression/Regression Percent.

This number, highlighted in green/red [option to turn on/off, turning off provides a more realistic/hardcore experience that makes it harder to determine if a player is regressing] numbers the percent of a player's progression that they have achieved. Each player has their own unique "progression cap," but everyone will start at a different level, based on their CHL/overseas stats and ratings. Players won't fall into the same pattern (the pattern currently is the higher drafted, the higher stars), meaning there's a chance that a late-round player could have either a high cap (though he may struggle to reach it!) or good personal progression. Presuming implementation of training into Be A GM 17, considering the fact that it's in Be A Pro 16, the Progression percent is something that a team can have a positive/negative effect on, rather than the current style of player being pre-locked into a path.

More advantages, and ways this reflects real life, to think about:

• Players have their own unique max value and will progress toward that at their own pace—some may hit close to their max progression cap, some may not (think "bust").
  
• The above reflects real life—teams may let a player slip because they think he'll have a tough time in the league and won't be able to improve much; yet teams may also reach for a player because they think they can develop him into a great player.
  
• Despite players growing at their own pace, you as a GM can affect their growth [more than the current play them in their correct role]—playing a guy in the AHL for a while may improve their overall progression, but they may have a lower progression ceiling due to the lesser competition. By contrast, you may place a player in the NHL early, but they may struggle and have trouble hitting a higher part of their progression.
  
• Rather than the current draft decision of #stars>colour, you'll have a much more interesting choice (but you'll have to wait until part 3, scouting, to see why).

Let's look at two real prospects from last year's draft to get a more direct inkling of how this works, without spoiling too much.

Conor McDavid is a guy with one of the highest possible attribute caps and the highest Progression Percent number in the whole class, due to his insane CHL numbers. He's a guy you'll want to play in the NHL right away without fear of his development being slowed. Lawson Crouse, on the other hand, is a "reach" type player. He may have a high cap, but it's tough to make the call on a project like him, not being sure about how he'll progress. But the high cap may be too tantalizing to let slip out of the top 10.

To address BigDisk's point from above—as I said, I'm still tuning my formula so that there will be fewer top-level players. But there are a lot of players out there with the potential to be great, but few make it to that height. I've put that option in place, while at the same time ensuring that it's rare that a player will hit their high max cap and be a superstar.

Scouting, of course, plays an enormous role in this. That's up later in the week. If you have any questions, comments, concerns, please share!