Ian Watches Soccer with a Computer - The Computer's Favorite Plays of the Season
/By: Ian L
Animations by: Mike Imburgio
Now that the regular season is by all credible accounts over, it seems like a great time to peer back through the annals of history and see which plays the computer liked the best from the ENTIRE SEASON. Let us comb through literal months of historical footage dating back to like May, and remember some of the campaign’s most pedestrian moments that a computer thinks are the most impactful plays of the regular season.
As a word of warning, there are a LOT of Colorado Rapids involved sequences here. I cannot explain this. I will not explain this.
Receiving - Gyasi Zardes - +0.541 g+
If you’re one of the two or three people that have been reading this series each week, first of all, thank you. Second of all, I’m sorry. But mostly, you know what this play is going to look like. Jonathan Lewis (who we’ll be hearing more from later), has the ball here out on the left flank. He opts for a simple ball towards Diego Rubio. Rubio, has far grander designs on this play. He looks up and plays a perfect rainbow from heaven kinda pass right to the feet of Barrios. Michel Barrios is an extremely dangerous player when he’s got you isolated out there. Ask Nouhou. Ask other left backs who they don’t want to see with the ball right there, and his name won’t be far down the list. He’s so quick in these small moments, and this is his go-to move. A quick shoulder feint towards the inside, and the fraction of a second pause it gives Kyle Smith is enough to create space for a low cross. Zardes doesn’t seem to hit super cleanly here, but it winds up working out well nonetheless.
Orlando’s center backs appear to have this run under control until suddenly they very much do not. Carlos is letting Zardes stay ball side of him, and Jannson’s third eye just seems to blink at the wrong time as Zardes is able to just sidle in behind him and somehow direct this over the goal line. Could Orlando have done better? Obviously. Hindsight, and all the platitudes that accompany this concept apply here, but even if we give Jannson the benefit of the doubt and say that he’s giving Zardes some space to bait that pass from Barrios, he has to be ready to cut it out.
It’s not a particularly exciting goal, and again I’d give most of the credit for it to Barrios, but the computer likes what it likes, and it’s gonna like the heck out of a player receiving the ball in that position.
Fouling - Tim Parker - +0.12
This is another one where the limitations of the computer are glaring. This is an exceedingly stupid foul. This is one where you’re flirting with getting sent off, and it was completely unnecessary. I hate this computer sometimes. Tim Parker is not a player our computer thinks too highly of, and this is it’s most favorite thing he’s ever done, so that’s kind of amusing. I’m not sure who to pin the blame on here for this play even sprouting. Parker’s pass is a bit lazy and telegraphed, but Quintero definitely hasn’t clocked Paul Marie, and turns the wrong direction leaving the ball concerningly available for stealing. Full credit to Marie here who reads the whole sequence of events before it happens, and suddenly San Jose is starting to look threatening. Marie dumps the ball to Remedi, who plays an awful return pass. This pass has a worse chance of finding a target than I do of ever getting paid to write about soccer. But Tim Parker and the computer have something in common here: Neither of them are able to recognize this fact. The computer, because it is a bunch of lines of code, and possesses no human eyes and therefore cannot see where everybody not on the ball currently is, and Tim Parker, because, well, honestly I can’t even begin to hazard a guess.
This ball should have just wound up in Steve Clark’s hands and we all go on with our day, but instead San Jose have a free kick in an area considerably more dangerous than where they would have had the ball which is nowhere. Parker is never getting sent off here, because the ref fortunately can see this is anything but a clear goal-scoring opportunity but why even pick up a yellow here and give the official any reason to possibly send you off?
All in all, this is a dumb one from the computer, but let’s do try and see this how the computer does. The computer can’t see that Clark is just going to pick this up. All it sees is a sudden counter attack moving very quickly towards a dangerous area of the field, that this foul stopped it, and the ensuing FK isn’t historically super dangerous. Therefore, a good foul.
That Tim Parker gets any credit here for anything is laughable, but such are the foibles of a machine that literally lives by a code. I wonder if Parker was just frustrated and lashing out. Honestly, if I were his manager I’d prefer that to just giving away the foul here. Anyway, congratulations I guess to Tim Parker for committing the computer’s favorite foul of the year. We’ll call it the Chara D’Or or something. I’ll workshop it.
Shooting - Jonathan Lewis - +0.10
Well, look, you have to know at this point what you’re going to get here. We have a goal of the season here in Major League Soccer, but I want an award for just the most MLSy goal of the year. Instead of the Puskas we can call it like, I don’t know, the Wondolowski award. Something like that. Add it to the workshopping list. It’s not BAD per se, it’s just not exactly joga bonito either is it? I do want to be clear here. The Colorado Rapids players in this play do basically every single thing right, and there’s nothing wrong with scoring ugly goals.
This play starts with Yarborough playing a ball up the field. 6’5 Jackson Ragen and 5’10 Diego Rubio each leap to win it. Neither does, and the ball bounces kindly to Barrios. As discussed above, this is not a situation where you want Michel Barrios to have the ball if you are supporting Colorado’s opponents. I think Arreaga actually does the right thing here. He’s not over-committing to the challenge, he’s just trying to keep Barrios wide enough that he’ll try a low percentage cross. It might have worked too, but the dominos started falling for Seattle 10 seconds or so ago. The backpass to Yarbrough seems to trigger a press (of sorts). Ruidiaz and Morris rush to close down Yarbrough. Nouhou slides to his left and moves forward to discourage the easy pass out wide to Beitashour. Ragen moves towards the space Nouhou just vacated to keep and so he’s a little further out wide, and a little higher than he’d normally be. This is what you’re supposed to do with a back three, but all of that doesn’t matter if you just straight up miss the header, which he did. Now Ragen’s out of the play, and Nouhou isn’t going to get back in time to help, and Rubio is completely unmarked so Barrios has an overlap he can play. He plays it.
Rubio actually hits a very poor cross here and Frei easily gets a hand on it. Should Frei have been able to catch this? Probably? I don’t know, goalkeeping looks pretty hard to me. Either way, Frei’s deflection should have been enough but Yeimar is a bit torn on his responsibilities here. He definitely turns his head and sees Lewis behind him, but he also sees a 2v1 developing with Rubio and Barrios and opts to go over and try to deny the cross. It’s an ambitious decision that ultimately pays no dividends.
Had he stayed with Lewis, Frei’s deflection would have led to a pretty routine clearance for Yeimar, but instead he’s now running back across Frei and the GK’s deflection just sort of hits him in the chest. This is surely going to be an own goal, but Lewis has been tracking this play the whole time, and is standing more or less on top of the goal line. He gratefully guides the ball along the final half inch of its journey and hopefully got some sort of goal bonus for his efforts.
All Seattle players are petitioning for offside, but that’s not how the rule actually works.
Your shot of the season folks. Thanks computer. I hate it.
Dribbling - Michael Bradley - +0.61
If you had asked me to guess who had the computer’s favorite dribble of the season, I don’t know who I would have said, but I can guarantee you Michael Bradley wouldn’t have made my top 200, and yet here we are, and surprisingly, I feel like the computer probably isn’t far off here. It’s a great dribble! This is one of the first outings of new look Toronto. The Italians arrived. Oh hey! There’s Kaye! This team is routing Charlotte at home! They’ll surely never lose again! Well. Nevertheless, this is a great play. Bradley’s header is collected by Bronico, but Jonathan Friggin Osorio gives him no time on the ball, and muscles it off of him. He taps it to Kaye who first times it to Insigne who plays a sumptuous (I’m sorry for using the word sumptuous, but you run out of adjectives quickly in articles like these) back-heel into the path of Bradley. The elder statesmen of US Soccer is probably near the bottom of the list of players I would have thought I’d want to have the ball here, but the old man proves me quite wrong indeed.
Bradley collects the ball well, shields Walkes off of it, and then with a sudden burst of speed that NOBODY including Christian Fuchs saw coming, just blows through the last line of defense, while still somehow maintaining close control of the ball before hitting a perfect dink over the helpless goalkeeper. THIS is the magic of the FA Cup, and it’s a MLS regular season game.
It may or may not surprise you to learn that the computer absolutely adores the work of one Michael Bradley. He’s the fourth best defensive midfielder in g+ per 96 (minimum 4000 minutes - he has TWENTY THREE thousand minutes ftr) in our all time dataset. If you think that means the computer clearly doesn’t know how to evaluate DMs, then just take a look at the guys around him. Atuesta, Joao Paolo, Keaton Parks, Osvaldo Alonso, Medunjanin, Dax, Chara…
It’s not just from his younger years either. The computer liked him better than any Defensive Midfielder this year as well. One thing we see with older players, is that as long as they continue to produce when they’re on the ball, the computer will love them because the computer cannot see them when they don’t have the ball. Goals added is very much (and is intended to be) an on-ball measurement, and that old man can still pass, and apparently dribble.
Passing - Xavier Arreaga - +0.36 / Patryk Klimala - + 0.33
We’re going to do two passes today because it’s a good object lesson for trying to understand the computer's perception.
The first one is from Xavier Arreaga, and it is just an absolutely horrid pass that only manages to rack up g+ because Danny Wilson just sort of… Look, I’m honestly not sure how to accurately describe what in the world happened to Danny Wilson here. The upshot is that rather than cutting the pass out, he falls over and it bounces over his head right into the path of a grateful Jordan Morris. Morris should have scored here, but his chip lacks conviction and Lalas Abubakar does a great job to back and cut it out. Now, if you were able to go back and digitally remove Danny Wilson from the play, (which is fair imo, because Danny Wilson basically removed himself from the play anyway) then you’ve got an exquisite pass that hits Jordan Morris after a perfectly timed run. That’s what the computer sees.
Or is it?
If you think that it’s crazy to award Arreaga the highest g+ pass value of the season because Danny Wilson fell over and the computer didn’t know about it, well buckle up and check out that animated gif.
Arreaga was awarded the highest g+ pass value of the season on a pass the computer very willingly acknowledges he did not even complete. Not only that, the computer clocked Danny Wilson too. He gets docked for losing the ball here. The computer saw ALL OF THIS and still just said, yeah none of that matters. Ball was in bad spot for scoring, now ball is in great spot for scoring. Whether Arreaga hit it where/how he meant to could not be less relevant to the computer. I’m obviously annoyed at the computer, but I would never in a million years have guessed it would pick up on all of that, so in a way, I’ve got to give my computer friend some kudos here.
Since that pass kind of sucked, let’s look at the second highest one:
This is pure uncut Red Bulls g+ cocaine. This goal is an entire proof-of-concept of the Red Bull ethos. Turn the ball over in midfield and go real real fast towards your opponents goal. Does Klimala foul Marshall-Rutty in the build-up? It sure looks like it. Does the computer care? It does not, and neither does the official. Shaffelberg is very quick to start making Klimala’s life difficult, and completely eliminates a pass to Yearwood as a viable option. It doesn’t matter. Klimala withstands heavy physical pressure long enough to somehow push the ball to Morgan who has nothing but daylight in front of him. He finishes with aplomb for his third of the day. It’s a really good goal, but kind of a weird choice for the best (or second best in this case) pass of the season.
When I see the computer assign a good deal of value to something I don’t particularly think it should, I remind myself that I need to look at the play as the computer does and remove all of the surrounding players mentally and see how the players on the ball moved it. So let’s do that here too, and we’ve got a really simple 5 yard pass sideways. It’s actually a pretty uninteresting pass if you take the context out of it. So how come it knows that it’s actually a really good one.
The answer to this is actually really interesting: I honestly don’t know.
I’m going to guess it has something to do with the coding, maybe as a through-ball. I’m going to have to go ask Matty.
Matty Kullowatz:
Okay, so that little pass that looks like a through ball was NOT tagged as a through ball. The key is that some months ago we added a one-on-one indicator to all models, letting them know when a player was one-on-one.
At the moment of that pass, the xPV (expected pass value) was about 0.01 = 1%. The model sees a ball moving quickly upfield, early in the possession, in the middle of the field, but doesn't know just how juicy this possession looks because it doesn't realize all the defenders are out of position.
Once the pass is made, the next action in the raw data is the shot some 50 yards up the field, which is tagged by Opta as being 1-on-1. The model has learned that this is a juicy opportunity and scores the xPV at about 0.50 = 50%. By the way, we set the carry xPV in these instances to the same xPV as when the player takes the shot, effectively fixing the carry value at 0. Thus the pass and receipt get all the glory = 0.50 - 0.01 = 0.49 g+ value. xPass here was 69%, so you get 0.34 pass value and 0.15 reception value.
Hahaha math folks! What a thing!
Anyway, basically what he’s saying is that the computer likes this pass for two very important reasons:
1. It led to a 1v1 opportunity (which are good!)
2. It wasn’t a risky pass. The computer loves passes that create great opportunities without doing something that carries a big risk of losing possession. A couple of strange ones from the computer here, but fascinating all the same.
If you’re wondering why Matty isn’t just writing this thing every week, the answer should be fairly obvious from reading how he thinks and writes versus how I do: His time is considerably more valuable than mine.
Interrupting - Ranko Veselinovic - + 0.66
Here it is folks. The highest g+ play of the season for any category, and surprisingly, it isn’t a solid back post run. It’s a defensive clearance, and one of the highest order, but also maybe kind of fixing a mess of his own creation.
We begin with the ball just sort of rattling around the midfield a bit aimlessly. Ultimately, it ends up at Cubas’ feet, but Montero reads his pass and cuts it out nicely. He plays a lovely through ball to Ruidiaz who gets in behind, rounds Cropper, strikes true, annnnnd then Veselinovic comes out of nowhere and saves a certain goal.
Now, if we break soccer down to its most basic component parts, it’s a game of goals and stopping a certain one off the line is a pretty obviously valuable contribution whether you’re a computer or just a person. However, has Veselinovic created the very situation that made him have to do something heroic to stop a goal from being scored? Was this all a grand plan to try and snag this incredibly obscure distinction? Unfortunately, we can’t know for sure, but people are asking questions.
This is the view from overhead. Can you see where the problem is? It’s pretty subtle, so I’ll show you what I’m talking about.
If you squint you can see where I’ve circled a potential problem area for Vancouver here. Yeah Veselinovic isn’t exactly in an optimum position to defend this current attack, but hey the important thing is that he makes up for it! The fact that Raul Ruidiaz doesn’t wind up scoring from that position in the screenshot is pretty remarkable, and the computer’s most favorite play of the whole season.
The computer has some very unusual tastes.
Goals Added isn’t quite a perfect metric yet, but it’s done a fantastic job of highlighting plays and players that we may have just simply overlooked because the things they do seem unremarkable at first glance. We’re constantly discussing how the model works, and what it’s telling us, so you can likely expect to see some tweaks to the model in the future as we get better, more specific data.
Well, I think it’s time for me be finished with this whole venture. I can go home and do something else, but the computer will still be here, watching, tabulating, trying to be instructive . You will tweet things about its love for Maynor Figueroa, and it will not care. You will say that the computer can’t actually tell you anything your eyes don’t show you better with regards to defensive actions, and it will not care. It will continue acknowledging and appreciating the subtle value that players like Michael Bradley and Andreu Fontas bring, and it will share this information with you freely, because it is incapable of caring whether or not people agree. It does not care if players are young or old, foreign or domestic, DPs, or TAM. It isn’t even aware that these are concepts at all. It has no concerns for my narratives or rooting interests. It has more soccer information than I could ever hope to hold in my brain. It has taught me a lot about soccer, and I have taught it absolutely nothing. But I’m still smarter than the computer, because I don’t think that misplayed passes and dumb fouls are actually good. Thank you for reading.