The Computer Playbook Project

[Sven Draconian]

The goal here is simple: To create a competitive offline experience by fixing AI play calling. This includes both offense and defense. Obviously sliders and coaching tendencies also play a major role in AI performance, and will be discussed as a sidebar.

Artificial Intellegence has only come so far, and there are things that it can do very well in this game, and things it can't. The CPP is my attempt to emphasis what it does well, while retaining enough seperation in team "styles" to keep the game fresh.

This means eliminating plays like Read Option that the AI simply can not execute properly. It also means finding ways for the AI to use their best players. In short, it's finding ways for the AI to succeed.

So with my trusty Phil Steele and some energy drinks, lets get rolling.

[Sven Draconian]

Game 1:
Started a Dynasty as MICH, first game was against Western Michigan.

Created a pro-style playbook for Western. I believe it was 4 "ACE" formations (A bunch, normal, trey and a 2 TE "trey" as well). 3 Gun formations (Normal, 4 WR trips, Trey) and a couple of I formations. 130ish plays.

Most of the runs were simply ISO, Dive, Draw or Power O.. A couple counters. No toss sweeps, 1 stretch. Out of the gun they only had HB Off tackle and draw.

Passing wise I eliminted all of the C routes, post corners, corner posts. Then I eliminated all the rollouts to the left, and most of the plays with poor spacing. I eliminated most of the "all curls" or "spacing" types of plays. I cut out as many plays with the arrow route in the flat (Curl Flat, Z-spot, FB flat) and swing routes. Obviously some of the flat-routes had to stay in, but I cut out a few. I also cut out all screens from under center.

I also tried to eliminate most plays with auto-motion, since they give away the play.

What was left was most vertical routes and drags. Those are the plays the CPU throws best (IMO). Plays like Levels, Stick, Mesh, Drive and then the play-action routes. Coupled with 4-verts, a few dig routes and double posts.

The game played well. Western Mich definately looked like a pro-team in terms of play-calling. 3 receivers most of the time, went to the I about 6 times, mostly to run the ball, and were in the gun 30% or so.

They ran 34 times and threw 22.

QB was 14-22 for 202, with a pair of picks.
RB had 21 carries for 43 yards, with a long of just 10. He lost about 10 yards in 3 carries running from the gun, so he was a little more consistant than his stats look.

I won the game 35-14, a fairly realistic score. I could have put up 42 but I let the clock run out in the end. Considering the ratings difference between the teams, the game played pretty solid.

Two things stood out right away:
-1) I recorded 10 sacks, obviously an absurdly high number. CPU pass blocking needs to go up.
-2) The CPU can't run screens even from the gun, and they really like to call them. They probably attempted 6 screens, were sacked twice, threw an incomplete-while-being-hit, and caught 3 for 10 yards combined.

[Sven Draconian]

Game 2:

Against Notre Dame (home), teams with fairly even ratings. I bumped up the pass blocking for the CPU.

Created a shotgun book with limited QB runs (only a few draws out of empty sets). No screens other than a pair of WR screens and a couple bubbles. Same style of passing plays as before, lots of crossing routes, levels, mesh, drive, stick. Avoided the complicated routes, tried to keep the RB in as much as possible.

I also assigned the CPU the 3-4 playbook I use myself. A lot fewer sets (I believe 5 of the 3-4 sets, plus the 3-3-5 stack) with all the unsound blitzes removed (blitzing 6 or more, leaving only 4 guys in man coverage) and the worthless blitzes removed (DB's coming from too far away, evenly distributed 5 man rushes).


Result was a solid 20-14 win for myself, even t hough I was sandbagging a little at points.

Stats were 181 yards for the CPU, 299 for me. I had 50 runs for 227 yards (option offense) while the CPU only managed 27 yards on 19 carries. CPU was 8-15 through the air for 153. No turnovers for either team. I only had 1 sacks, and relatively little pressure.

Thoughts after 2 games:
- The shotgun run game is still horrific. The CPU ran off-tackle/sweep about 10 times this game, and lost close to 20 yards combined. The draws were great, very productive. I'm going to test the HB gut play and see if the CPU can use that.

- I've been playing with CPU accuracy at 40, I might bump that back up to 50. The CPU is lobbing too many balls.

[Sven Draconian]

One other note. Using my custom 3-4 playbook, the CPU regularly came out in 3-deep DIME defense against my 2 flexbone formations (no issue against my 3 "I" formations). I'm removing that defense from the CPU's playbook in the future.

The CPU loves to do a handful of (stupid) things if you give it the choice.
-It will call HB screen 5 times a game if it's in their playbook.
- It will come out in a prevent-style defense against passing formations

Even though I want the CPU to be able to use those plays, it will use them far to frequently if they have access. They will not be included in future books.

[JPowser]

I dig this thread! Mind throwing in the formations and some plays from the custom playbooks? :D

[Sven Draconian]

I dig this thread! Mind throwing in the formations and some plays from the custom playbooks? :D

Once I start to finalize the books ill post them. Right now Im still throwing stuff around. Dont feel like posting 100 plus plays if Im going to change a lot. This is just a journal of sorts, hopefully others will use it as well.

[JPowser]

I'm interested so you'll have at least one fan of these I'll be following for sure!

[Sven Draconian]

Played around in practice mode quite a bit today trying to fix up the CPU run game from the gun. Found some interesting nuggets.

The biggest issue with the gun run game is the angle of attack vs the angle of the block. Since the OL doesn't have a true reach block/hook block animation, it just tries to drive the man. When the CPU runs the off-tackle/stretch/sweep plays from the gun, the DL can easily dis-engage the animation and make the tackle. As humans, we can manipulate the angle to setup the block how we want it, or simply cut the play back inside. The AI always wants to attack the assigned hole, even if the defender is right in it.

So the only viable plays for the CPU gun game are the "Base" (where the guard pulls and leads) and counter (same blocking, just with a counter step) as well as the draws. This is because the OT/TE does a pass set, and the RB's angle takes him inside of that block.

In previous editions, these plays took too long to develop. However, it appears that a hitch-step has been taken out (or something) and the plays hit much smoother. Counter is especially effective, since the LBs will usually bite on the counter step.

[Sven Draconian]

Game 3: Vs Eastern Michigan

Not too much value can be put into games with such a mis-match (Michigan with a B+/A-, EMU being the worst team in the game), but there was some promising things.

Since they are listed as a "One Back" O I just let them play with my CPU Pro book (same was WMU in Game 1), with the changes to the gun running game mentioned above.

Won 49-0 (EMU skipped a FG chance late). Lots of penalties on EMU (3 or 4 false starts, a few holdings) hurt them a few times on decent possession. They also had 2 turnovers.

Offensively it was a cake walk, Denard Robinson had 156 rushing and 3 longish TD's on the option, Gardner had a 50 some yard run in junk time. Total of 36 caries for 321 for me on the ground, but Eastern has horrendous defensive speed.

Eastern had a moderately effective rushing offense. 54 yards gained by their best back (albeit, on 21 caries), but he was consistantly getting 3-4 yards. A few negative caries cost him in terms of YPC, and he never broke a big one. But, their rushing game was pretty effective....especially considering how awful they are.

Passing was meh. 10-21 for 174 and a pick. I had 5 sacks (which killed the net rushing total).

Some thoughts:
- Stretch (from under center) will likely be getting removed for the same reason as the Gun-run game doesn't work. Might try and sub in a HB toss instead, but those have historically been ugly.

- The CPU has been struggling on the deep ball, letting receivers out-run their arm and taking too many sacks.

Mostly, I need to get in a game against tougher competition again and see if the offense moves the ball better. I'm trying to avoid moving the sliders around too much until then.

[Sven Draconian]

Game 4: Vs San Diego State

Another test for my CPU Pro-style playbook. Even though SDSU was a 3-3-5 defense, I just had them use my 3-4 playbook. Overall it was a very good game, I won 20-17 in the last minute. Based on ratings Michigan is the better team, but both teams are in the B range.

Yards were 225 vs CPU, 263 for me. (I had 34 car, 175yd on the ground). Neither team had a turnover.

Passing wise the CPU was 17-21 for 189, even though I had a couple of bad plays on the stick or else that is 15-21, possibly with a pick, and one fewer TD. I recorded 4 sacks (mostly on deep balls being held to long).

Their leading back is a 90 ovr, good player, but only gained 66 yards on 20 carries. Again, the running game for the CPU was consistant, but lacking big runs. I would say the CPU should have been sniffing 100 yards on that many carries.

Thoughts:
-Overall the book is playing well. The CPU is going 4-verts too often, and can't really throw it, so it is taking too many sacks and letting the WR run too deep, resulting in lobs/jump balls in traffic. Might remove it.

- I'm going to raise the CPU rushing sliders just a hair. RIght now the CPU is getting a solid 3 yards a carry on average, but it needs to get closer to 5 to compensate for the amount of TFL it takes.

- So far I'm pretty happy with the HUM off vs CPU def using my custom books. Not nearly as many gimmies.

[Sven Draconian]

Game 5: Minnesotta

Big success this game. Tweaked the sliders a bit (upping CPU run block and RB ability, dropped HUM Run defense one click more. Had a great game with very good results.

Another win, at home, vs a lesser opponent (20-17 this time). The CPU ran for 108 yards on 24 carries, including their #1 HB having 13 carries for 74 yards. Passing wise the CPU managed 10-13 for 147, still to high percentage wise, but the yardage per attempts is solid. CPU threw 2 picks, I threw one.

Human stats were 45 carries for 229, and 3-4 passing for 118.

Thoughts:
-Scoring is too low because 7 minutes (with accel clock) isn't getting enough plays. I ran 50 as a human, CPU only ran 36. A typical college game should have in the 120 range from scrimmage , so we are running 50% fewer plays. This is turning your typical 31-24 type of scores into these 21-14 I've been getting.

I think I'll try 9 minute quarters. Since I'm an option team I should be on the lower side of plays per game, but definately still clearing 100 a game.

- CPU 43 D book is playing nice. I'm getting consistant looks from the defense so I can actually make some play calls and have an idea of what I'll be looking at.

- I think the CPU is adapting during dynasty. They were using a lot of BEAR and under looks, which the CPU was not doing early.

[Sven Draconian]

Game 6: At Northwestern

Real interesting game. Created a new Spread Option playbook for this game. Only 6 formations (5 wide, Normal, 3 WR, Trips, Trey, Normal HB wk) since they had the best QB run plays. Took out the read options and any form of tiriple option and went with the same style of passing plays that have been working in the other books (PA, crossing routes, stick, levels, mesh, drive, shallow cross).

Bumped the quarters up to 9 minutes.

Yardage was 408 (me), 383 for Northwestern.
Rushing was 48-309 (me). 35-172 Northwestern.
Passing was 7-15 99yd (me), 20-39 211 Northwestern.
First Downs 11 (me), 19 Northwestern.

Yardage was fairly even, Northwern rushed the ball extremely well (172 yards). However, I won 38-7 on the strength of 7 turnovers, 6 of them interceptions (1 pick 6).

Turnovers have been decent in all previous games, so Im assuming this was just an abberation.

Thoughts:
- CPU used 5 wides a bit much.
-QB scrambles killed me. I was playing soft zone D most of the game, and the CPU ran the ball well.
-Could not establish the option, CPU D was all over Denard.
- 9 minute quarters is too long. CPU had 78 plays, I had 58. Too many combined considering we had 75 rushing attempts between us. Also, takes a tick long.

[Sven Draconian]

For those keeping track at home:
I've created 3 offensive books for the CPU at this point:

Pro-Style (Ace, I, Gun...3 of each IIRC)
Spread Option (6 formations, featuring a lot of QB runs)
And what I've labeled Air Raid for simplicity, but it's really just shotgun formations without any QB runs.

Still need to create a power run O (teams like Wisconsin, Iowa, Alabama) and a true "Multiple" offense, which will be the pro-style book with the addition of a few QB runs (Boise State, Ohio State ect...).

Defensively I created a 34 and 43 book. I won't be making a 3-3 book, it's not common enough at the major schools to occupy a slot, and I'm not sure the CPU can run it anyway. I may still make a 4-2-5 book, I need to poke around inside and decide if it's worth a slot.

[Sven Draconian]

Meant to add that I'll post a formation + plays from each book soon. Generally, the books contain the same plays from multiple formations...and posting 600+ individual plays seems tedious.

[Sven Draconian]

Game 6: At Michigan State

Nice and tight win for me, 24-21. MSU was driving late (at my own 40) when they had a big fumble. They outgained me in yards (358-309), and they ran for 157 (HB Edwin Baker, 28-105). MSU went 10-15 in the air with 1 pick. I won the turnover battle 2:1, which is what ultimately gave me the win.

If it wasn't for a few scrambles by Denard I would have lost this game, but he had a pair of really long scrambles on 3rd downs.

Teams were fairly even ratings wise (Michigan being slightly higher rated), but the game was at MSU. Probably need to toughen up my sliders still, but a good game never the less.

[Sven Draconian]

Game 7: Vs Purdue

Took my first loss, 27-17. Tweaked the sliders right before the game (turned down Hum RBA is the big one) and I played like sh*t on offense. I only managed 230 yards, 39-139 on the ground and 6-20 in the air. The CPU outgained me by 100 yards, going 33-264 on the ground and 9-14 for 100 in the air.

Bolden (HB, Purdue) had 3 long runs...including a 99 yarder... which skewed the stats. I also took 7 sacks as the user (RBA nuked my bootleg/sprint out passing attack). Denard ended up with 21 att, -2 yards.

Thoughts:
-Might remove speed option from CPU book, it's been nothing but a TKFL in both games I've seen it, and they call it a bunch.

-CPU does not call HB counter or base enough.

[Sven Draconian]

Here is how my CPU Pro book is looking:
Formations:
Ace
-Bunch swap
Normal
Trey Open
Twin TE
Y Trips

I Form
Normal
Tight
Twins
Y-Trips

Shotgun
Normal Fex Wk.
Trips HB wk
Y trips HB wk

Here are the plays in I Form Normal
-Runs
HB Counter
Iso Weak
HB Stretch
HB Lead Draw
HB Iso
HB Power O

-Pass
Ark Fades
Slants
PA Streaks
PA USC Y-Post
HB Choice
PA Power O
Bench
PA Comebacks
PA DBL Go's
Cross In
PA Clearout

In ACE Normal
-Runs
HB Dive
HB Draw
HB Iso
HB Draw (again)
HB Stretch

-Pass
Slants
PA Boot
Four Verticals
PA Ctr Waggle
PA Slot Crosses
TE Post
Flanker Dig

That should give you a decent idea what the types of plays I'm using. The playbooks generally consist of the same plays from several formations. I still plan on pairing these down a little to try and get the CPU to use it's #1 WR more and use more PA, but that is some framework.

[Sven Draconian]

Game 8: At Iowa

Fairly even ratings, on the road. Toughened up the sliders again just before this one. Took a decent loss, 24-10. I was playing pretty loose, just testing things out. I had 4 turnovers (2 fumbles, 2 picks) while they had 0. I did outgain them yardage wise (176 to 171). I only managed 38 caries for 121 yards, but 10-17 for 155 in the air. Iowa ran for 192 on 32 carries, went 4-8 for 79 in the air.

Iowa's Adam Robinson ran for 80 yards (16 car) before going down. Their backup had 92 yards on 13, with 76 of those being after contact. (Robinson only had 19 yards after contact).

Thoughts:
-All in all, another solid game. Dropping hum RBA has really hurt my run-around-with-denard passing game, which is good.
- Created a new book for this game, a run-heavy book. Iowa played just like they should have, pounded me with the ball and milked the clock with the lead.
- Iowa felt like they were running an offense, lots of repeated plays and looks. They lined up in Weak I twins 10 times or so, ran a little ISO, a toss, and a few passes. Mixed them up well.
-I included some auto-motion in this book, since a lot of good running plays use it. Helped bring some continuity.

This is probably my last game in this test dynasty.

[Sven Draconian]

Here are the sliders I'm using as of now. Playing on All American

Cat............HUM/CPU
QB Acc.........30/40
Pass Blk........50/60
Catching.......50/50
RB Ability......25/60
Run Blk.........30/65
Coverage......40/60
Pass Rush.....45/55
INT..............40/50
Rush D..........35/60
Tackling........35/55

Still trying to find the right balance between HUM Tackling/CPU Run Ability and CPU Rush D vs Hum Run Block. Trying to create consistancy (carries between 2 yards and 10 yards) rather than tackles for loss followed by 20 yard runs.

[JPowser]

:up: thank you this project is looking food still!

[willowyboot500]

just read on another forum that custom playbooks are for users only. is this wrong cause i cannot get cp to stick with cpu controlled teams.

[Sven Draconian]

just read on another forum that custom playbooks are for users only. is this wrong cause i cannot get cp to stick with cpu controlled teams.

use 2 controllers. assign 2nd player to CPU. once you kickoff to start game, when the returner catches pause it and go to select side and move player 2 to the middle. Typed from my phone, sorry if its choppy sounding.

[willowyboot500]

gotcha. so during dynasty sim games, which playbook will computer use. can i make oklahoma a wishbone team or nebraska option I, kinda 1970's offenses for cpu teams to use and they use them.

[Sven Draconian]

Sure can willow. Just create a playbook with what you want the CPU to use. Assign it before the game (using 2nd controller), wait until the kickoff, move the 2nd controller to neutral and the CPU will use the created playbook.

I would always use Michigan as the base for the created books. I have never experinced the 0 play glitch using the Michigan playbook as the base.

[Sven Draconian]

Odd game today. I created a cupcake Fordham Rams squad for my real dynasty. First game is against Army. Since I run an I-based option book (3 I-formations, a pair of flexbone formations, a Gun and an Ace look for 2-minute drill/3rd and 10), I just had the CPU use that book.

They only called 3 or 4 unique plays the entire game. Almost every play was run out of the goalline formation, either ISO, off-tackle or Power O (plus, 1 "toss") and then 4-verts out of Gun Normal (a play that can't be deleted from Michigans playbook). I lost 41-21, due to special teams and the fact that the game was a gross mismatch (60ovr to mid 70s)...but I can't figure out why the CPU would only use those plays.

The CPU ended up running for 311 yards (IIRC) on 41 carries.Just lined up and pounded the ball up the middle.

[Sven Draconian]

Took a video of one of the playbooks, easier than posting the 100 some plays. It won't upload to Youtube (stupid phone), but assuming it work I'll post all playbooks in video format soon.

Had a nice game tonight. Played FCS East in my 3rd game with Fordham (cupcake roster). Lost 36-17. Lost my starting QB early in the game, which crushed my offense, and my D was awful. Gave up over 500 yards, (190ish rushing). The CPU threw 20-29 with over 300 yards.

Feel happy with the results of my sliders + playbook combo. Now I just need to step my game up some more.