Alex Andorra 00:00:02

Let me show you how to be a good Bayesian. Today, I am thrilled to be joined by Ravi Ramineni, a pioneer in the world of soccer analytics and the architect behind the data driven success of the Seattle Sounders. With a rich background that bridges large scale data analysis at Microsoft and deep rooted passion for soccer. From his days growing up in India, Ravi has uniquely shaped how analytics are applied in sports today. In our discussion, Ravi takes us behind the scenes of building an athlete management system and a cutting edge scouting and recruitment platform. At the Sounders, he shares insights into how data analytics has revolutionized training decisions, injury prevention, and player acquisitions, emphasizing the critical role of accurate data interpretation. To avoid costly misjudgments in player evaluations, Ravi and I also tackle the complex issue of evaluating coaching effectiveness, the quest for more predictive metrics in soccer, the enhanced need for collaboration between modelers and club decision makers, and the future possibilities with advanced tracking data. This is learning Vision Statistics episode 116, recorded June 26, 2024.

Ravi Ramineni 00:01:26

Let me show.

Alex Andorra 00:01:27

You how to be a good Bayesian and change your predictions after taking information in. And if you're thinking I'll be less than amazing, let's adjust those expectations. What's a Bayesian is someone who cares about evidence welcome to learning patient statistics, a podcast about bayesian inference, the methods, the projects, and the people who make it possible. I'm your host, Alexandora. You can follow me on Twitter alex endora likethecountry for any info about the show. Learnbased.com is lambastobe show notes becoming a corporate sponsor unlocking bayesian merch, supporting the show on Patreon. Everything is in there. That's learnbasedats.com. if you're interested in one on one mentorship, online courses, or statistical consulting, feel free to reach out and book a call at topmate IO. Alex and Dora, see you around folks, and best patient wishes to you all. And if today's discussion sparked ideas for your business, well, our team at Pymclabs can help bring them to life. Check us out at Pimc dash labs.com Ravi Ramineni welcome to learning bayesian statistics.

Ravi Ramineni 00:02:44

Hey Alex, thanks for having me on.

Alex Andorra 00:02:47

Yeah, thank you so much for taking the time. You are always a man in demand, so I am extremely honored to have you here. And thank you to Patrick Ward for putting us in contact. Patrick was here actually on lbs, episode 111. For people who want to have the background about Patrick, we talked about us, football and the fantastic work he's been doing at Seattle for ten years now. Well, today we are going to talk about sport two. We're going to talk about sport that's called football, too, in a lot of countries, but in the US, it's known as soccer, so it's a different sport. It's not the football that Patrick loves so much. But, yeah. Like, before that, though. Yeah. Can you tell us what you're doing nowadays, Ravi, and maybe briefly how you ended up working on that? Because if people want more background about how you came to work on football and so on, I will put in the show notes, a great episode you did with Ted Knudsen on his podcast, the transfer flow that's going to be in the show notes. So feel free to keep it brief, Ravi, or tell stuff that you didn't tell yet to Ted so that it's not repetitive for you.

Ravi Ramineni 00:04:14

No, I mean, first of all, you know, thanks to Patrick, amazing guy. I miss him in Seattle. At least meet him a few times a season and talk shop and all that. Yeah. So right now, in this moment, I am trying to narrow down my options for my next project. I just exited a startup that I co founded called Source Football. Before that, I was working at Seattle Sounders for about ten years where I, I built first their athlete management system, then scouting and recruitment platform. Before that I was at Microsoft. So I was a program manager. But a lot of my work involved analyzing huge amounts of data that is collected on search engines, the Bing search engine, and tried to understand whether a user is satisfied with a click or nothing. You know, very different than. Very different than football. But it is still data analysis, but it's a different data. Yeah. So that's my rough background. So right now I'm just, I have a lot of proposals, a few that I sent out, a few that people have sent out to me. I'm trying to figure out based on what I want in the next few years of my life, what I want to work with. I'll still be in football, soccer, but figuring out, finally narrowing down those choices, hopefully in the next couple of weeks, I'll have something to say, what I'm doing.

Alex Andorra 00:05:57

Yeah. So thanks. That was a great and brief overview. So it's perfect. And again, I recommend the episode with Ted because you go in details into how the hell did you transition between Microsoft and working on search engines and now working on football. So for people interested in a senior career path, and I really love that because I've had a lot of people on the show, ended up with a blend of determination and randomness in what they are doing right now. And I always love this kind of path. So definitely recommend the episode with Ted and I'll stop the teasing here. What we are interesting here is, well, first, you talked about an interesting nugget in your presentation right now, which was, it sounds like it was very different what I was doing at Microsoft from what I'm doing nowadays. But it's actually not that much, and we'll get back to that, because I know you have great ideas and points to make about that. But first, what we all leave on here on the show is patient stats. So, ravi, do you even know what bayesian stats are? Have you been introduced to them in your curriculum? Do you use them even in your work nowadays? What's the state of that for you?

Ravi Ramineni 00:07:33

So I was introduced to Bayes theorem a long time ago, I would say towards the end of my high school, and my favorite subject was probability and statistics in college. So, yeah, I was introduced to that. But Bayesian statistics themselves, that's something I haven't heard or I haven't really dug into, or just try to see what it is about until I read this book, the signal and the noise by Nate Silver. So, yeah, so that's just to answer your question. Like, you know, I know the base theorem, I have used it a lot, but never, never kind of like learned that on the statistics side. I think a lot of my statistics background was just the frequentist, the usual stuff. And. Yeah, like, I know the problems of the, how I tried to learn it. You know, try to learn. I think Stan is the language, right. That does. There was brief moments of my role at the Sounders where I was trying to learn to see how I could use it, because there are certain problems that once I understood what Bayesian, how bayesian statistics work, that they would address a certain problems in a better way than like, let's say traditional, more frequentist approach. But I never got the time, or maybe I never had the, like enough of, I had the starting problem. I would get to a point where I hit a pep plateau, but I didn't have time to seek out help, and I, because that was not part of my day job. So it went like it was a hard one to kind of get over. But that's something that I would love to explore because I have a lot of questions and ideas on how to solve certain problems in football using bayesian approach.

Alex Andorra 00:09:43

Yeah. So if you have questions today, feel free to, to ask me that. That would be fun in the episode.

Ravi Ramineni 00:09:53

Maybe another time for that, because I think there will be like, yeah, but I'm sure I will reach out to you on that. I'll take you up on the offer.

Alex Andorra 00:10:01

Yeah. I mean, for sure. That's always fun. And as people know, I'm a football fan, so, you know, always happy to advance the cause.

Ravi Ramineni 00:10:12

Who's your favorite, favorite team?

Alex Andorra 00:10:15

Yeah. So that's a big problem because I am, unfortunately, a fan of Paris since I'm like five years old, and that's very hard to be a fan of Paris. You know, I'd really, really rather be a fan of Real Madrid, for instance, you know, or a smaller club, but Paris is really a roller coaster and since I'm like five, I've. I've known everything with that club. Like, we've been extremely good at the national and the european level and then we were really near relegation at some point and we were literally saved by a goal at the very last minute of the very last game of the season by one of the worst players of the whole season. That was like a really good illustration of football and its randomness.

Ravi Ramineni 00:11:12

Exactly, yeah.

Alex Andorra 00:11:13

And, well, today, yeah, it's in. And nowadays really hard to be a fan of Paris again. I can develop on that, but later.

Ravi Ramineni 00:11:23

Okay. So.

Alex Andorra 00:11:24

Yeah, but basically I was actually gonna ask you if you had, if you have personally a club, because I think it's harder, at least in my experience. I don't really like having emotional attachment, actually, to a club. Watching Paris game, for me, is actually an agony, whereas watching any other team play, I love it. I love the game and any other sport. Right now I'm working in baseball and watching the Marlins and any other team. It's a real pleasure because I'm not completely overwhelmed by emotions. So I'm curious how it is for you.

Ravi Ramineni 00:12:08

Yeah, I would say that I. I'm a big fan of vill, which is a small team in Spain. The yellow submarine.

Alex Andorra 00:12:17

Yeah.

Ravi Ramineni 00:12:18

I would say, like, before I started working in football, I was. I would watch those games very differently. I think as I worked more and more years in football, I think I've gotten a little less attached to any football game, to be. To be quite honest. I think that I've started developing that distance and also I watch the games a little differently. But for the longest time, watching real games, even while working in football, that was a different experience for me than watching any other team. But I completely understand how you feel about. Especially if you are. If you're born and brought up near that club and that's club, that's your club from so young. So then it's. It makes a lot of sense.

Alex Andorra 00:13:07

Yeah, yeah, yeah. It's like you don't have a choice, you know? It's like you have to still love the club, even though, like, through thick and thin, it's like you're married to the club no matter what happens.

Ravi Ramineni 00:13:19

Yeah.

Alex Andorra 00:13:19

And I've tried, you know, I've tried. I've been disillusioned by the club so many times. I've tried. At some point, I was like, no, I'm done. I'm not gonna watch the games anymore. And I could never do that. And how do you like that? Actually being a bit more detached, I can definitely understand, and that's good for me. I'm happy to hear that because maybe it will do the same for me, but, yeah, how do you like that? Because I can definitely envision some people being like, oh, yeah, no, I actually love that emotional attachment. I wouldn't like to be less attached.

Ravi Ramineni 00:13:53

I think it's more that I don't have the time to invest, to be more emotionally attached. There was a time in my life where I didn't have to worry about so many things in football. I was only worried about watching one team, or maybe a few teams, but primarily interested in one team. Now, through the job, I am living and breathing football every day. Like, I won't say 24/7 but most of my work time goes in, goes in football. So I think that now I don't have that. Okay? Now, out of this, I need to carve out an extra time for an extra, like an emotional experience. That's the. That's the thing. And the other thing is, as you see more and more players and more and more football, they are not that different. Right? As much as you want to think, obviously different uniforms and different ideologies and different styles of play, but it is still football at the end. And so I think those are things I think I, you know, obviously these are things I reflect upon because I think a lot of the changes happen subconsciously. You don't know that you are. You are actually detaching yourself until you get to a point suddenly one day when you reflect and say, like, wow, this is not how I used to watch VRL games. I used to have a ritual that I probably has done that for a good part, more than ten years, is I listen to the local radio and have the tv on with the tv muted, and that's how I listen, that's how I enjoy those games. And the video was always, like, 20 seconds ahead because the guys in the city, in the stadium, and, you know, and there was directly streamed, and obviously, Tv has some delays, so especially with streaming platforms and all that. So that was the way I enjoyed those games. And every now and then, I still do that, like, if I have a chance, you know, the games kick off at morning time when I'm going for a run, I just switch on the game and put it. Put my headphones and just. Just run for 2 hours just listening to the game. But I don't do that as often as I used to do because I used to do, in the past 38 league games, plus, plus, like, whatever, european games, I would do that for 50 times a year, but now I probably do it, like, ten times a year. So that's what I mean by kind of building that distance.

Alex Andorra 00:16:24

Yeah, yeah, that makes sense. And that's really funny, because I do the same with Paris games when I'm like, yeah, yeah. I mean, these. These season, for instance, I was sometimes in Argentina, and so I would watch the game with the argentinian tv, but I would put the french radio, thanks to the time difference, I have the podcasts coming from France, so I don't listen, I don't watch the result before, and so I use the podcast, but I can make the podcast at the same time as the video, so I don't have the delay.

Ravi Ramineni 00:17:06

I see. Okay. The funny thing is that, like, the. The broadcast local broadcasts are in Valencian, which is Catalan, which is a slightly different version of Catalan. I speak fluent Spanish, but I also understand, like, because of this, I learned Valencian pretty well. You know, like, I can understand it. I can't speak because I don't. I never practiced, but. But I can understand it very well. Like, you know, I really like a lot of phrases. It's really close. So, yeah, it's a fun way to learn a language, too, I think. For me, at least.

Alex Andorra 00:17:39

For me. Yeah, yeah, no, definitely. I mean, that's also how I learn languages. Most of the time, I go to the country, and then I'm forced to speak. Yeah, you don't have a choice. I think you should go to France now, because, like, if you know Spanish and a bit of catalan, then French, it's gonna be super easy for you.

Ravi Ramineni 00:17:58

Yeah, I tried to learn when I was in India. Like, I used to go. I used to have friends that I don't know if you heard of this organization called Alliance Francaise. Oh, yeah. It's like a cultural organization that's around the world. They do french things, french cultural things, and also do classes. So I took a class and I had friends that. That would come to my city in India and teach French for two semesters and go, like, that's part of their college work. So I used to hang out a lot of those, France. And, you know, I. There was a point in time I was getting to a point of making small sentences, but I kind of lost it after that. So, yeah, as you said, if I get the chance to live in France for six months, I can definitely pick up a good amount.

Alex Andorra 00:18:43

Yeah, yeah, no, for sure. I mean, well, you're welcome anytime. Let me know. I'll connect you. And actually, that's funny. That makes me think, you know, when you were saying, in the end, when you watch a lot of games, it's kind of like all the teams are a bit different, but also similar. So first, the nerd in me is like, okay, hierarchical model. It's like, hierarchical models just express that. But then also the traveler in me, I've traveled in a lot of countries, and by doing that, actually had the same reflection, you know, both countries in the end, you know, and it's like, that's. That's why now when I end up hearing about very nationalistic people, and unfortunately, that's the case right now in France, where we have a lot of these nationalistic vibes, unfortunately. Like, this is really weird to me because I'm like, you really think the country's, like, really something really special? Like a special snowflake, but not that much. If you travel, you'll see it. In the end, it's still people. People are really similar, and. Yeah, so I have some experience.

Ravi Ramineni 00:19:58

Problems are similar. I think it's. Everyone is trying to, like, at a global level, everyone is trying to get to the same place, I think, you know, in terms of basic needs, and. And then, you know, once you get to the basic needs and then maybe a little bit long, you know, just trying to secure their future. Like, to me, like, those two would cover 99% of all what people are trying to do. And so. And it's the same in every country now. Every country, just because of where they are and what their history is, has certain level of resources and certain issues and certain geographical things. So when you're born in a country, it's nothing. You know, it's almost like it's sort of whether you're born in the US or you're born in India or Lebanon or, you know, France, that changes so much. Right, but there is no but beyond that, I think it's all the same. And all the differences we've created I think, you know, that's, that's something that we've kind of created along the way. But, yeah, it's not when, when somebody is born really, they don't really know who they are.

Alex Andorra 00:21:01

Yeah, no, exactly. That's what I tell say all the time. And I'm like, you know, well, I was lucky to be born in France and it gives me a great passport, but, you know, I didn't do anything for that. Just random luck, you know, so I don't really understand. And then, yeah, when you travel, you actually talk to the, to the immigrants, you know, that then people who didn't travel a lot say they come and they steal jobs and so on. But you, when you talk to people, you see they don't come to France to do these evil stuff. It's just they want to live and live better most of the time. That's that. But, yeah, well, thank you for that parenthesis. Definitely appreciate it. That makes me think about my, my first career was in political science. Of course. It's still, these topics are still dear to my heart, even though I don't work on that anymore. But so getting back to football, and I'm not surprised in the end that we ended up talking about social thing because football is also a very social and emotional endeavor. But so to go back to one of your experiences, the one at the Seattle Sounders, I'm curious because here you basically participated in founding the analytics department over there. What were some of the key goals that you had in mind and you wanted to achieve with this?

Ravi Ramineni 00:22:42

Good question. Different levels to it. First of all, the first few years they hired me to build like an athlete management system using fitness data, heart rate data, GPS data. That was what I was hired to do in the beginning because I was hired by head of performance who is actually a great friend of Patrick, also David Tenney. And he. So that was my first job and I didn't have any experience in exercise science or sports science or anything like that. But obviously I had experience building systems database that are based on a database and has a web interface that will take in some data and put out some reports. So what I was tasked to do was, well, we just bought GPS and heart rate for the team. Can you take this and help us make sense of it? Because we want to inform our decisions about how players are training and what we need to do. If they overtrained the day before, should we take some measures next day? Or if they didn't get enough training stimuli, stimulus the day before, should they get more the next day. So they, so they had the domain knowledge of what is a good training day, what is a bad training day, what is an overload, what is an underload. And I was then tasked to take these GPS raw data and build a system that would then capture that. So a lot of so on a purely technical side, and this was back in 2013. So technically the first system I built was a standalone SQL server that is in our, kind of in our, I think it's like, it's basically an on prem server. And my choice of language at the time was on the front end was c sharp. And so I built like a web interface using c sharp and then have the SQL server in the backend with stored procs and, you know, rdbms data. And so we get this data from, and also we didn't have the API from, from the GPS or heart rate at the time. Now there are APIs. So I had to take these data like CSVs. Basically, all these softwares in sports, they come with their own little kind of software. All these kind of technologies come with their own software. So from that software you can download a CSV and you take the CSV and then upload it into my system so that you can have this global, where I have, we used to do a lot of tests for players to see their readiness and fatigue levels. Put all this information together to come up with a continuous story of how this player has come here. Obviously, the goal is always to prevent injuries from happening and to know when a player is at risk and try to take remedial action before the injury happens. Now, one of the funny things there is that a lot of people talk, you know, like there was a time when people use the word injury prediction models. A problem is that your data is never clean enough because anytime you know a player is at risk, you're going to intervene and that data is useless to predict from that point. Right? So a lot of sports science data suffers from this problem. And even I think, you know, in general, like, I think it just happens in every, every sphere, is that when you know something is going to go bad, you're always changing the inputs to, because you don't wait to see something to go bad before you. You're not experimenting. Like, I give you an example going back to my previous job in Bing. Now it's not previous. Now it's like four generations ago. Like we used to do experiments, we changed the color of the font, or let's say we change search results page. So when you search for a query you know, you get a list of results like blue links. What we used to do is that, well, let's change the font for font size from 13 px to 14 px. So then we'll then serve this new, new experience, which is 14 px for 2% of the users, but the rest of the 98% get the same old experience. We run this experiment, we call this experiment, we run this for two weeks or whatever time, knowing how much data we need. We'll run that for that period of time and then analyze the data, compare. We have a lot of metrics to say whether which one is better than the other. And then we would analyze. Now there you're doing a pure experiment, like a b testing, but a lot of times in real life you can't do a b testing because there is a real cost to not acting on when you know something, right? So anyway, yeah, so my first job was to build that. That was what I did the first two years and then create reports, create just capturing everything that happened in the training. And I did a lot of things of like collect my own data to classify what are the different types of training drills. And so I had to build like a little, like a small kind of nomenclature language that classifies different drills as different things. So that when I go into the system I know what is the drill, how many players were involved, who were involved, what is the size of the pitch that was used, what are the rules? Like, it was goals, no goals, goalkeepers, whatever. So yeah, I did all that in the first couple of years. And then after that I was tasked to do what I really wanted to do in football was to do the scouting and recruitment platform to analyze players, analyze games. So then I started doing that. So the goal there always for us with was number one or two, like you wanna, obviously, like this is, this is captain obvious sign. Good players avoid making bad mistakes. But I think that, but that with the salary cap league, very little money, if you make a mistake, especially with the designated players or the players that make the most money, you can only have two or three of those. And so you end up having to be really careful. So one of the things that I always describe my process as is that I'm always, I can live with as many false negatives as possible, but I can't afford a false positive. So the job ends up being, you analyze 1000 good players or a thousand players, you reject 100 good players and you probably, and you want to sign a good player. So I am okay skewing towards missing out on some good players and some high potential, as long as I don't make a mistake. I think that was a, that was kind of for the, for that group of, for that, for the money where we have to spend a lot of money. So that's how I would describe our process was. So it was a lot more like a, you know, like, I guess we call it, you know, band pass filter type thing where we might even, basically we might eliminate some high potential because it comes with high risk, but we always eliminate the lower section. And so you always kind of end up with a steady, kind of a above average type of decision. That's probably where we wanted to live most of the time. Yeah. So that would be a rough way that in terms of technology itself, similar technologies at the beginning use the same database, same front end, but for the visualizations, we used tableau at beginning and then we then added some of the models we built in r and some of the models were, what I would do is that I would experiment in r and then, or Python, sometimes python, very little in the beginning. And then basically once the model is ready, I would just code it into SQL stored proc because that would make it faster to surface. And I'm also not a great software engineer, so that was not my strongest suite. So I had to do things that would get me quickest to where I'm good at or what I really want to do was analyze players and things like that. So that would be our structure. And then as we got into, I think, 19, at the end of 19, we moved from the Stone Age to cloud. We moved off of the on Prem SQL server onto a cloud based system where we migrated everything to the cloud, which was a huge endeavor to move everything, especially when you don't have a lot of resources. And it's like, hey, what did you do last three months? Well, we just squared one. You won't see any result of it because we, we're going to be the same whether it's coming from the cloud or coming from a SQL server on Prem. So that was kind of. And later on we added more. We got bigger data sets like tracking data to do, like more custom metrics, line breaks, and better models for expected goals. And. Yeah, so as we got tracking data and, you know, those are towards more recent times, you know, so that's roughly the what we built there.

Alex Andorra 00:32:52

Damn. Yeah, that's, that's a lot of work and thank you for working, walking us through all that. And I think there is very interesting nuggets of wisdom and information here for listeners, even if they are not working in football. So let me ask you the question in a more general way. From this experience you had, what would you say is something that's applicable to someone who is doing data science in the organization? And they have to build this kind of infrastructure, and then they have to use it and help people use it in the organization. Which principles can you draw from that? And would say people would be able to apply them even in an organization outside of football?

Ravi Ramineni 00:33:52

Yeah, I look at this as very much the core of this is making a decision. Everything in life is a decision. Whether you want to go to a restaurant, which restaurant, that's a decision. Now, that's a very low leverage, smaller decision. You don't really repent much whether you go to one or the other. But when you're talking about a decision that involves 50% of the salary budget of a team, we're trying to talk to this player that is really good. Got to be one of the best players in the league, but it's going to cost 50% of our salary budget. So when you're making that decision, that's a big decision. Like, you can't get that wrong, because if you get that wrong, you set the team back for a few years. So I see all of this as the process, is that there are two things, the decision and the process. The process you use to make the decision, analytics is a tool that can help the process of making decisions, that can help you get around some of the biases that humans have and will make it easier for you to make a good decision. But I can give you another. I can give you an example that I've used in the past. So if you're. Let's take a car. You're driving a cardinal now, any car these days, it has the rear view mirror and the side view mirrors, and you are even subconsciously using them when you are changing the lane, or in general, when you are always seeing who is around you, just so that you have the sense of if you see something suddenly happen in front of you, you need to know which way to swerve or you want to change a lane. That if you have a side view mirror, that's really easy. If you want to go left, there is the mirror. You look at the mirror and then you can turn. And there is obviously, there's still having these. There is still a blind spot, meaning that you still couldn't be perfect. You still want to turn your neck and see. That's what the good practice is. Imagine driving a car without those side view mirror and the rear view mirror, you can still drive the car, you can still change lanes, but you will have to turn your neck more often back to see what's there, to decide to see what's there. And there is more likelihood that you miss something that was there or you miss a blind spot. So it puts a lot more effort on you to make sure that you don't make a mistake. I see analytics as these tools that if the rear view mirror is right in front of you, it's so easy to look at it. You just kind of lift your eyebrow a little and you see it. And so, you know, like, okay, there is somebody behind me, whatever. So I look at it that way. So I think the whole process of making a decision, you want to arrive at a good decision. And this process, there are various tools you can use, and one of those tools is analytics. And what it does is that it takes some of the emotions out of the equation, some of the inherent confirmation bias and other types of biases that we all have that take some of those out, or at least point us towards them to act upon them. That's how I see analytics. Now, we talked about a car earlier, we were talking about football, signing a player that will make half of your salary, budget, how to use up. It's the same thing, I think, like any, any field, like, ultimately, you want to make good decisions. And another thing about, you know, decisions is that, and you want to make this process, a repeatable process, to continually make those good decisions, like, and because on a given basis, any organization is making dozens and if not dozens, hundreds of decisions on a daily, weekly basis. So you want to have a process that guides you. This is how we make these things here. Now, that doesn't mean that it would be always perfect, just like even having rear view, side view mirrors, it still be an issue. Sometimes there is a blind spot and you. You shift the lane and there might be an accident, or you come really close to an accident, but that doesn't mean that, but it still helped you avoid a lot more. And as I said earlier about injuries, the moment you do an intervention, the fact that you prevented an injury from happening, or the fact that you prevented an accident from happening, means that there is no trace or evidence of that. But that's something that is part of it. And that's how I look at analytics. And when I see what I do at a more fundamental level is help teams get better, install a process that they can repeat, and make good decision. One on top of the other.

Alex Andorra 00:39:10

Yeah, that's actually something you've talked about in a very interesting conference at stats bomb last year that I've put in the show notes about compounding of decisions where. And I think that's a very good point because there is a lot of also, not necessarily from modelers, I would say, but from people who consume the models. An idea that the models can be, you know, sudden game changers, kind of like how things happen in a, in a movie, right, where it's like, oh, there was a before and an after and it was very clear and black and white, but most of the time it's much more the accumulation of small decisions that were all going in the right direction that in the end steered you in the path you wanted to, ended up being at. You know, it's like if you want to get in shape, it's way better to go to the gym very regularly and do 1 hour than go one day per week and do 3 hours. So this idea of compounding is, I think, very interesting. And I'm actually curious from your different experiences because you've worked with a lot of clubs and that means a lot of managing teams, so very different people in front of you. How do you think, has analytics transformed the decision making process within football, particularly in the areas you know best? So player signing, players, game strategy, and if you have an example of how analytics led to a successful player acquisition, that'd be awesome.

Ravi Ramineni 00:41:05

I think that analytics had a big impact on the sport of soccer. I mean, I think if you talk about baseball, I think you could easily say right now everyone is like, if there's just a lot of attempts to just score home runs, right. In baseball, just like in basketball, either they take a three or a two pointer close to the basket because that's come through analytics. Similarly, in football, if you look at the long range shot with the player taking a 30 shot from 30 yards out, that's almost gone now, or it's a very low percentage, because I think I would say that's direct impact of analytics where we've kind of beaten it into the heads, saying that taking a shot where it's a 2% probability to score is not a good, is not a good decision. And again, going back to, like, the decisions, like, even on the field, we're always talking about player decisions. Like, did the player make the right decision to pass the ball here or did he have a better option? Right. But anyway, like, coming back to analytics itself, impact, I think that I would. That's the most clearest sign of analytics impact on football is something that you could see if you count the percentage of shots from outside the box, which would be 18 yards out in, say, 2010, and you compare that in 2015, and you compare that same number in take any league in the world and compare that in 2020 and 24, you would see a steady decline in that number, probably down, I would say, 15, 20% from where it was. And that's due to just this becoming more mainstream. So that's a great example of this. In terms of my personal examples, I think that one of the things, as I said, analytics is one tool in the toolbox of making these decisions. And one of my personal, like, the way I try to operate or I try to message everywhere where I've been, is that don't treat analytics as a separate. Oh, this is like, this is a separate entity. I think that in terms of decision making, it's another source of information. So it's another tool to help you make that decision. I wanted that as integrated as possible, where on a daily basis, my ideal scenario, my ideal club would be on a daily basis, there is data, or analytics is used almost involuntarily. It's just a part of the culture, it's just that, well, we're trying to sign, or let's say, take a different example. We're looking to, this player's contract is expiring in 18 months. What do we want to do with him? Right. So there are certain principles. Okay, what is his age now? What is his age in 18 months? So where is he in the career phase? Whether he's passed his peak or he's going to pass to speak, how long of a contract we can give him and what would be the market value? Like, those are kind of table stakes. Like, you have to have all that information, and that's all driven from the. That's all powered by analytics or analytics department. But that's something that you almost, like, expect it and that you follow that process, it becomes part of it. That's how I see it. So I find it really hard to separate out and say that this was an analytics decision, because it's never like that. Like, if you go inside any organization, football or otherwise, not one piece of information is responsible for a decision or a huge decision. Like, you know, a big decision that, especially ones that are successful, I think they always have multiple people involved, multiple departments, multiple information. So it's always like a committee based approach. And then there is also something I've heard from, in another talk, where when you have more than one person making the decision. Like when you have a committee, what it does is that it normally means that humans. Humans are not good at. Are very bad at recognizing their own biases, but they're really good at pointing out others biases. So when you have a committee that is a good one, not like, you know, where there is trust and where a group of people that work together and they know each other well, then you know that when you discuss a decision, then it becomes more. You point out all the flaws in how you came there, and it usually ends up being what is left over after you beat it up, as they call it, and you end up. Usually end up with a good decision there. So I think I will see it that way. I think that in my time at the Sounders and even afterwards, I've tried to do that. And there are moments where obviously, when something goes right, you feel, okay, definitely, I want this to happen. But again, I think at the end of the day, I always say that the only people that deserve any credit for success is the players themselves, because they are the ones who are playing, you know, everybody else is. I don't know. I just feel like it's hard to separate that process. I think that's. That's the reality.

Alex Andorra 00:47:12

Yeah. Yeah, that's a very good point. Does that make it hard then, to sometimes having to say something that's based on the model or, like, your analysis of the data that's not going in the direction of that. The other parts of the organization are wanting to hear. What's your experience on that? Basically, what I'm asking here is, how do you deal with resistance?

Ravi Ramineni 00:47:52

Yeah. So I think this is a. This is a difficult question to answer just because a lot of times, if it is strangers, like, let's say I just joined a club and I don't know anybody there. I'm just. And obviously I know them, who they are, they know who I am, but we don't really have a relation. It will be harder to go into a meeting and say, you guys are looking at this wrong. This is my analysis. I think we should do this other thing, or do we sign this other player? I think the way to do that would be, I think it will take time. Obviously, when you start out, it's hard. It's almost impossible. I think you want to build that trust in the beginning. And the way to build that trust would be to first listen and understand where they're coming from. And what is the idea here? What are we trying to do? What is the decision we are trying to make and why and learn about the background of it and then have smaller conversations with individuals or maybe small group of individuals about what you can do for them. And you have to convince them how what you do will help them do their jobs better because ultimately that's the buy in, right? The buy in is ultimately people will use your stuff if they believe that it's useful for them. It's as simple as that. So you want to build that buy in. And the way you build that buy in is, again, like show that you understand or you know what you're talking about and that you understand what their problem is and where they're coming from and, and then build a solution. And over a period of time, let's say whatever, like maybe it's a, take six months or a year, I think you, you will get that comfort and they will get that comfort with you where they are comfortable, you challenging them, you are comfortable being told that you're wrong and have that argument. At the end of the day, your job, let's say if you're an entry level analyst or maybe you just have one or two years experience or you haven't really built that much of a cache of credibility and all that, you want to make sure that at least they listen to you and they consider what you said because everybody, you know, the trust gets built at its own pace. You're not going to dictate that because that's the two person that has to happen from both sides. And so, and again, like, you know, you, you say your piece and you go back and let's say they didn't agree with you and they did the thing that you said won't work and it doesn't work, then that's a counterfactual of like, okay, well, you know, he said that to us. But then there is, I mean, it's not as simple, I admit that, because especially in sports and football where careers are short term, you make a bad decision six months later you may not have a job. And that's also where the resistance comes from. Because when, when there is, when there is so much short termism in organizations, people will want to do it their own way. And I can't fault that. If you put yourself in the situation saying if I get this wrong, I'm going to get fired, then will you listen to people around you or will you want to do like, okay, if I'm going to do this, I'm going to do it my way? I think that I know I'm in this job for a reason. I'm going to do it this way. I think that's what happens a lot of time, and that gets kind of mixed with that. People are resistant to change, or people are resistant to analytics or a different type of information, but I think there is, like, we need to tease those things out.

Alex Andorra 00:52:01

Yeah. Yeah. That makes me think back to a lot of my personal experiences, because I'm often the modeler who comes into the organization, and sometimes we have to give some hard truth to the teams we're working with. And also, it makes me think to something that Paul Saban said in episode 108, where I asked him a similar question, and I think he made a very good point by saying, a great way to build a trust with the people you're working with at the very beginning, as you were saying, is actually to try and confirm their priors at the beginning. So if you have a model that you can show that it confirms their prior in something that's actually true with some degree of certainty, then you're building trust here because they will see you're not here to, you know, destroy what they've had done. You're here to help them make better decisions, as you were saying. Because in the end, that's the. The name of the game and we're building models. But in the end, you can have the best model. If it's not used, well, it's useless even if it's the best. So you need that trust internally. And that question. Yeah. How to beat that trust. I think it's a. It's a great way to do it. To do it here.

Ravi Ramineni 00:53:37

Yeah, no, I totally agree. I think that's something that we tend to do, even with models that we built, just to get a first sense of whether this is doing the right thing. If you're trying to list the ten best fullbacks in the world, and your model determines ranks the fullbacks in the world, look at the top ten, and then they generally tend to confirm the board view. The first thing you check is, well, whether seven or eight of these make sense to be in the top ten, like, because that would be pretty. There would be a lot of consensus around that. So that's, as you said, that's the same kind of logic there where you confirm the priors and you saying that, oh, at a fundamental level, this is what this model is supposed to do. And if you can get that confirmation that, yeah, we all think that this is doing the right thing at a fundamental level, then that's the first step, then you can go from there. And I think the other part is that I think with football analytics, there is a lot of limitations to the data. We've been talking about models and data. There's a lot of limitations. You can't answer a lot of questions, a lot of questions around in football surrounding what happens off the ball, because a lot of the data available is just event data, which is changing. But still today, overwhelming amount of data available is event data. So you can only answer certain questions. You can't answer a lot of things around how the player plays off the ball, even with the tracking data. Tracking data is just a two dimensional xy location on the field. So you don't really know whether he's facing the goal or it is back to the goal. So that pose, that information is not widely available today. So then you're still approximating and you can still find ways around to get a proxy type of thing. But what I'm saying is there's still a lot of questions you can't answer. And another way to build trust, or it's more of a, I would say it's like a counterintuitive way, I would say is be ready to admit when you can't answer a question that I don't know, like, I think that, or my data or the data we have or the models I have cannot answer that question because a lot of times I do see the tendency wherever people want to say yes to everything that, you know, I can answer every question or try to extend themselves over, extend to answer a question, even though, you know, deep inside that probably that's just not, there's no signal there. Yeah, I think that's very important, too, because the quickest way to lose trust is when you get found out that way.

Alex Andorra 00:56:40

Yeah, I mean, definitely that's something I say a lot on the podcast also. So now listeners are going to be like, oh, great, dad, again. But I think, yeah, something is indeed, you have to be able, as the modeler, to, you know, communicate what the model can say, but almost even more importantly, also what it cannot say. And that, to your point, is extremely important, I find, because that also goes a long way towards showing that, you know, you're not here to show that what people have done before is completely stupid and now you're going to fix everything. But on the contrary, that you're building on the shoulders of giants, if you want. And in that, yes, here, here is what we can do. But that no, would have to do something else or we would have to modify, modify the model or we would have to collect more data. Something I find usually that's the case here, is when people have causal questions, but the model or the data, most of the time, can only answer correlate correlation questions. Yes, here we know that happens at the same time. Doesn't mean one thing causes the other. If we want to do that, we can. We can, but we need to run that experiment, ask the players to wear that ring while they sleep, you know, so that we can measure their sleep. And then, yes, we can answer that question if you want, but right now we cannot. And that, I think, is very interesting.

Ravi Ramineni 00:58:21

And so, like a lot of times I used to have these conversations with. This is all at Seattle with the assistant coaches and, you know, who have a little bit more time to discuss concepts and things. After training, I would have this discussion, and I would talk about correlations and say, well, from the data, I see this is what's happening. And then they would then go on and explain what they'll ask me, what do you know what's happening tactically there? I would say, no, I don't know, because I want to learn. And then they will go on and talk about, like, well, this is happening because we told the player to do this, x or y. And the way football works is that when I move this player here, it creates a chain reaction of reactions from the opponent team, and that opens up this space. And so there is a lot of nuance. Now, I don't have data for all of that. And that's the problem. You quickly get to a point where you reach the limits of what you can explain with the data you have, and then you get to a point where. Yeah, so I agree. I think that you have to understand the causality and a lot of time correlation. I mean, it's a cliche, but it is true that a lot of times correlation is nothing. Causation.

Alex Andorra 00:59:49

Yeah, yeah. I mean, for sure. And sports is definitely a great endeavor where you are most of the time interested in causality. And I think it's a great playground for nerds like us because most of the time people are interested in causality, but you have to do experiments or quasi experiments or be smarter about how you go about the model. And I find that fascinating, for sure, because myself, for instance, at the Marlins, I have access to extremely talented people who know so much about baseball, much more than I will ever know. And they also know about niche. Like, they did a PhD in strength conditioning, and so they know about that intersection about, how do you make a player fitter and bigger. But at the same time, how will that impact his performance at bat, for instance? And that is the kind of stuff that I don't know about, but that's just gold as the modeler because I go talk to them and that helps me get where I could not go before because I could not even have imagined that the data I'm seeing right now in my computer is actually a result of that change they made. So, yeah, as you were saying, I think here it's also a very good way to also make other people in the organization stakeholders of your models, because then if they had a stake in your model and seen the model having better results, that means your model is going to be better, but also the model is going to be more used, which is what you want.

Ravi Ramineni 01:01:36

Yeah, I think that I was going to point that out earlier is one of the ways to build trust, is work together to build metrics or specific metrics. And there are various techniques here, like you can do some of what they want. A lot of times you also encounter this in the analytics world is where a decision maker wants to just confirm his decision, so he wants some data to prove or confirm what he was thinking was right. So that in itself is not, in isolation, is not a bad thing. You always want to confirm whether what you're doing is right or not. But the motive of that could be different. Like sometimes if it is just to kind of a box ticking exercise, then it's not good. But if it's in general as like a checks and balances, I don't think it's bad, but you could use that as a way to get an in saying, well, you know, these are the metrics that confirm with what you say, but there is these other metrics that don't confirm and this is what those metrics are about. And I think that's another way to start building that trust. But I think universally if you work together, collaborate on the models and on the, on the metrics, they will get used more.

Alex Andorra 01:02:57

Yeah. And I think that's where also the Bayesian framework is extremely interesting because you need priors, so you need to talk to domain experts and it's just naturally included in the modeling workflow because then even before running anything on the model, you're going to have to define a priors and so you're going to have to talk to the experts. And I find that super interesting to also when you can, what I do often is talk to a domain expert, tell them that. Okay, so your priors are these and these plug that into the model structure I currently have, generate prior samples and then show a few graphs to them that they can understand about the thing we're studying and then ask them how reasonable that that looks to them. And that's also a good way for them to constrain their priors, to calibrate them and actually sometime to update their priors. Or also a lot of the time for me to change the model structure because my model is too wide in a way, it allows for too many scenarios. Most of the time it's this that's actually interesting to me, that a lot of people, when they start using patient stats, they fear that their priors are too constraining. But most of the time, because you have a lot of parameters in the model, what ends up happening is that actually the model is too permissive and you're allowing scenarios that are way too extreme in reality. So it's a funny thing that happens. Interesting. So we're running a bit low on time here. I don't want to take too much of your time. Can I keep you for a few minutes still?

Ravi Ramineni 01:04:45

Sure, yeah.

Alex Andorra 01:04:46

Awesome. Yeah, I have, well, actually, a question that popped into my mind when you were saying that, well, good data is actually a challenge sometimes in football. Still. What kind of data would you like right now to answer a question you really care about these days in football? And right now you can't because you don't have the data. Like, you know, if you had a wish to make to a genie right now and they told you, okay, I am the data football genie. What data do you want for which problem?

Ravi Ramineni 01:05:26

Pretty. Pretty. Yeah, I think. Well, I want. I think right now what we have is like, I'll give you a quick summary of what data we have. We have what we call event data, which is just recording of all the events on the ball and some metadata like Xy location, stuff like that. So that's about 3000 rows per match. And that data is available the most widest. So you could get that data for second division, Ukraine or third division, Korea. Like, you could get that between different data providers where you could get across the world. Just the coverage is wipe. The next dataset is there's optical tracking data, which is the camera systems installed in the stadium. Those are very much like, you only get the data for your leak because it's a league wide deal and it's like a lot of restrictive, very expensive also. And then there is like the intermediate thing that has come on in the last five to seven years called the broadcast tracking, where they're just taking the broadcast video and using computer vision to generate tracking data for on screen, and have, like, algorithms to predict, or at least, at least for physical data purposes, like just predict where the players are when they're off screen, just to get their kind of distances and things like that. So these are the three major data sets. I think one that's coming up soon is a Hawkeye data set, which has pose. They say, I have not looked at it. Obviously, the data size is much bigger. I think that that's probably where the biggest thing to measure in, because we're all measuring proxies of like in football. Like my coach, one of my coaches at Seattle, he had this definition of what does a play in a football look like? So there is execution of the play, meaning you are passing the ball, vision or decision making, where you are deciding what to do with the ball, and perception of perceiving what's around you. So for a player to decide, he's looking around what's around him and then deciding and then making the action, which is the actual execution. Most of the metrics we have today just measure execution. So we don't measure, really measure decision making. We do proxies. Even with proxies, we're really far behind in measuring. Did the player make this decision? Unless you go ask the player, really, we don't know yet. So I think that what I would want is I would want to be able to quantify the decision making, because at the end of the day, you want to make good decisions. Going back to what we've talked a lot in this podcast, it's about the decisions, and the more good decisions stacked on top of the other, the better. And so, how do we make players take better decisions? Can we give them feedback? What are good decisions? So I would want a data set to be able to measure that directly, not proxies, but that would probably mean I need. I'll tell you how complex that would be. Is that from the line of sight of the player, what can he see? The field when he has the ball, when I have the ball, when I'm scanning, what do I see? I need that information to be able to, or if I can record what a player saw before passing the ball or before deciding to shoot, that would give me the real information to answer that question. Now we're talking. We're going from trying to answer the causal parts of it, trying to address the causal parts. Obviously, that looks a little impossible right now if you think about it. But that's probably where in that direction. Anything that will lead me in that direction, get me two steps forward. I would want that.

Alex Andorra 01:09:50

Yeah, that makes sense. I love that. That would be super fun. I mean, then the type of models you can do, all that also just increases exponentially. Yeah, yeah.

Ravi Ramineni 01:10:03

I mean, like, it probably involves, like, you have to put something on a player shirt or something, like, obviously wearables, like cameras these days. You can put them anywhere. Like, can't be intrusive, but, but, yeah, but there is a lot of different things. Like, I know some clubs have done some experiments with, like, players wearing goggles and there are these kind of 360 systems where there are some exercises and the whole thing is, like, fitted with cameras that is recording a lot of things, but still, that's not the real thing. That's just a simulated lab version of it. I still don't think that comes close to capturing the real stuff.

Alex Andorra 01:10:44

Yeah, yeah, for sure. But damn that, I'm sure that's gonna happen at some time and that's gonna. Yeah, then the kind of models we can make would be incredible. Something. I would be curious. I am curious, actually, about myself, and that may be because right now my mind is really working on a project. I have to translate war from baseball to football, actually. And so my mind is really in the, you know, the replacement level mindset. And what does, what is the contribution of a player in comparison to the replacement level? Something I'd be very curious about is doing that for coaches and that, like, I don't know how to do that yet. I don't even know if that would be possible at some point, but I would be very curious to have the data to do that.

Ravi Ramineni 01:11:41

Yeah, I think it's a little difficult project because I've had to do some work around that in the recent times, but it's really hard to separate the coach effect. Yes. Yeah, it is very hard. Like, it's just, you know, I think that. I think, I think from my experience of trying to do it and probably mostly, mostly ending up with something that I'm not 100% satisfied with. I think most coaches are just like most players or replacement level, meaning that, you know, I know. So just that, you know, there's not that much difference between one and the other if you give them the same rosters. But, but then there's definitely, like, the outliers that are really good and, like, I think the one, the one thing where, you know, I've heard this before is that a good coach will improve the team a little bit. Like above average coach will improve the team by some amount but like a bad coach can really take it down. So like. But then there's like this whole middle that probably is not that. That different.

Alex Andorra 01:13:07

Yeah. Yeah. I mean that's a fascinating question to me also because, well that's a causal question. Right. So. Yeah, and that's, it's so hard.

Ravi Ramineni 01:13:17

And I, you know, this is just again, like I don't think this is a hundred percent causal inference that I was able to make but I think, you know, it's, it's the kind of educated guess or type of things that you can do. But again, that's, that's not causal. That's again like is you're going to heuristics. You know, it's there. They're what they are.

Alex Andorra 01:13:43

Yeah. I mean, and that's fascinating to me because I think it's even harder than for players because players, at least they are on the field. You have a lot of years of data. They can stay in a club for years, which is not the case of a lot of managers. Less and less actually. And usually the managers who stay in place the longest are considered to be the best. So you already have assembling bias here. But also, especially now, the best, the managers consider the best also the ones who get the best players. It's like, you know, there's multiple bias all over the place. Yeah.

Ravi Ramineni 01:14:28

Yes. And then it's like. Yeah. It's a different magnitude. Yeah. It's just hard that, that. Yeah. You know if, if a coach does well with a smaller group he's rarely stays there for long enough to actually be able to. To do that.

Alex Andorra 01:14:44

Yeah. Yeah. So yeah, I mean, look like Chevy Alonso, Athenae Leverkusen. It's rare and I'm not sure he'll stay there for longer. You know, like at least next year but then probably he'll hold jump to something else to much more.

Ravi Ramineni 01:15:02

The other thing is also like, you know, obviously I can give you another name. Kieran McKenna who is at Ipswich Town. Two promotions in a row from League one to Championship to Championship to Premier League in two years. He's still there. So he's going to manage in the Premier League. So I'm really fascinated to see that as an experiment. But even then, even if you know somebody's good it's hard to say that what is the, you know, if you put him in a different setup with a different group of players like is that like okay, if he wants to play the same way then you have to maybe change half the squad to at a new place and it still won't be the exact same players. And so there is a lot there, like, I think in terms of manager search and manager identifying who's a good manager.

Alex Andorra 01:15:52

Yeah, yeah. But I think it's very important too, because not only I find that very interesting, but also clubs spend a lot of money on managers, so it's not a decision you want to take lightly because also if you do, like, the contracts can also be very long. I know, right now, I think Julian Naglesman, for instance, he's still paid, I think, by Bayern Munchen, even though he's the manager of the german national team. And then it's like they still have to pay the contract because otherwise it's going to cost them millions to let the whole staff go. So really, if you don't make the right decision, it can be very hard to bounce back afterwards.

Ravi Ramineni 01:16:39

Yeah, I mean, that's not an uncommon thing. A lot of clubs, I know clubs that. That are still paying their previous two managers.

Alex Andorra 01:16:46

Yeah.

Ravi Ramineni 01:16:47

You know, and that. Yeah, it's not very uncommon, which is pretty bad. I think there's a lot of that in football.

Alex Andorra 01:16:52

Yeah, yeah, yeah, yeah. I mean, and that's weird. It's just when I heard that was possible, I was like, how's that possible? It's just. It's incredible. So, I mean, there's no. I don't think there is enough data right now to do really good job with that. But I think probably it's something that clubs are going to look into after. After some point.

Ravi Ramineni 01:17:14

Yeah, once they're already looking into it, you know, obviously the more progressive ones are already looking into it.

Alex Andorra 01:17:20

Yeah.

Ravi Ramineni 01:17:20

It's just that I think just even having. You don't need a perfect answer. Just having that discussion and going through the process and getting to a point where you get to a point where you reach a dead end, like, yeah, okay, I can answer question three out of the ten questions, but that is still better than zero out of ten and you make the decision in a little bit more like, you know, without. Without that information. So. Yeah, for sure. Like, I think there's ways to go there.

Alex Andorra 01:17:46

Yeah, true. That's actually a very good point that also I make a lot to my students. It's look for a good enough model and good enough priors and not. Not the perfect model, not the perfect priors because they don't exist. Right. Just get to the good enough point. If you can already help make a better decision with that, that's already very useful. Yeah, I mean, definitely, if I would have access. If the multiverse existed and I had access to it, I would definitely do a random sampling of the coaches very regularly. I would like. I'd be like, okay, all coaches get, I don't know, three years in the same club and then we just change at random np, random choice, you know, with the number of coaches and we just change everything. That would help a lot.

Ravi Ramineni 01:18:37

That would be a lot. I mean, I guess like there is maybe a scenario with AI in the future if you can do that type of simulation across coaches and across teams. And it's more. I mean, it's still a simulation, but maybe you can get, you know, I mean, that problem is like you have an issue with the fidelity of the data in soccer. So you don't really have that level of data to be able to do like a realistic simulations. So maybe, maybe that's the path. Like, you know, you. You get to a point where you can simulate that.

Alex Andorra 01:19:11

Yeah, yeah, yeah, no, for sure. Yeah, well, talk about something else. But yeah, it's like also. My point is it's also harder to define what's a successful coach, at least players. It's easiest. It's not easy, but easiest because it's like, well, the forwards has to score goals. A successful coach is not only a coach that wins stuff. So that's really also the hard thing.

Ravi Ramineni 01:19:38

Yeah.

Alex Andorra 01:19:41

Actually something I'm curious about is in episode 108, Paul Sabin, he said that the scientific mind he would have dinner with, which is like the last question I asked, every guest would be Virgil Carter, who is. I don't know if you know him, but he's us. Football player. Yeah, NFL player from the seventies. And he wrote a paper about expected points in football while he was playing in the NFL. That's so cool. And so I was wondering, but just, you know, a bit more generic. Did you have experience with players or coaches like that who were really nerdy and who were really willing to have that kind of analysis? This is just curiosity for me, not.

Ravi Ramineni 01:20:36

Not to that level. You know, I've definitely had coaches that are very interested in how I arrived at certain numbers and, you know, would talk about it a lot of times. It just. I think. I think a lot of times we had a good relation, but the coach, they always were looking at like, well, we want to prove you wrong type thing. But, but yeah, you know, but, but it was more like. But I had, you know, I have good relations with them so I was okay with that. Yeah, but never. Not, not to that level of nerdiness, for sure. Not to your question. I. I was like. I would say, like, I. I was going to give you two names. One of them probably not. Not a scientific mind, probably, but a really beautiful mind is, uh, you know, I come from India. I would love to have just a chat with Mahatma Gandhi. You know, that would be the one. I would probably. That was the first name I. I can think of. If I want to go scientific, I would want to talk to Srinivasa Ramanujan, the famous mathematician who only lived, like, 33 years, and his equations are still being like three. Are figuring out the proofs. Yeah, I would. He's actually from the south of India, you know, close to where I was born and brought up, so I would love to. I would love to, you know, hang out with him. I know he must have. Awesome. Yeah.

Alex Andorra 01:22:09

Oh, yeah, that must. Yeah, that would be an amazing dinner. Damn. And that's the first time it happens in the show that someone answers the last question before the last question. So. Thanks, Ravi. That's great. I still have first in that show after more than 100 episodes and. Yeah, actually. So, I mean, I could still ask you a lot of questions. I have so many in my mind. But let's be. Let's start closing out here, because otherwise I'm gonna. I'm gonna, you know, draw all your strength and you'll still have to work afterwards. Yeah. One last question before the very last question. Just in general, I'm curious to pick your brain about where you see the field of soccer analytics heading in the next years, and if there are emerging technologies or methodologies that you think will be very interesting, if not game changers.

Ravi Ramineni 01:23:06

I'd say a lot of work in the tracking data so that you get better tracking data, more fidelity. As I was talking about, the Hawkeye data set, stuff like that becoming more widely available right now. They only collect for one league, I think, and there's not much if you are only collecting for one league. It's really not that useful, because for player recruitment, you need that data across a wide spectrum. The problem is that it's never going to happen in soccer. That I would say never say never. But I'm sure we can come up with a technology at some point where the second division argentinian game, we could have Hawkeye data, like, as outlandish as it sounds today, maybe there are technologies down the road that could answer that, or that could do that for us, but until then, we'll have to probably settle with using the broadcast version of tracking data and the improvements in that, I guess the other thing is a lot of work will be done on the predictive metric side. I think a lot of metrics in soccer are really good at describing things. And I think predictive, and then I don't know much about prescriptive, but probably predictive is where a lot of work, I think a lot of progress to be made on that front. Predicting like, that's one of the biggest questions among many, but probably the big one is like how to predict, like the career progression of a young player and at what age are you going to have enough confidence in that prediction? Right? Is it 14? Is it 16? And part of the challenge of the problem is that development is not linear. And so then it's not like you come up with the model and it's not that easy. And then, you know, there is a lot of things that go into it that are not even in the control of the player. Where maybe, for example, like this happens a lot at different clubs in their academies, is that they have three play three, the three best prospects play the same position. So let's say it's age 16. And so what ends up happening is that one of them will have to, at least one of them will end up having to play like. There are multiple possibilities. One of the possibilities is that one of them playing slightly different role or not his best position. So that could decrease his kind of projection right there. Another could be that the player will leave and go to a different team where the setup is different. So that could decrease his potential to. And another could be that the team made the wrong choice of figuring out who should play his best position and who are the other two that are expendable. The team may have made a wrong choice and picked the one that probably didn't have really was not going to go anywhere. So all these things, I think. I think that's type of questions, you know, in terms of when I say predictive metrics and predictive things, I think I'm talking about things like that.

Alex Andorra 01:26:28

Yeah, or. Yeah, or like.

Ravi Ramineni 01:26:30

Or a player moving from a league to league to predict what will be his performance in the, in the, in the new league. Yeah, yeah. It's just like right now it's very much not. Not very reliable, those predictions.

Alex Andorra 01:26:42

Yeah, yeah. I mean, so first, the example you just gave about the three players that made me think about when PSG had the luxury of having both Ibrahimovic and Cavani and both were playing nine, but they could not because of how the team was structured. And then Cavani has to play on the left side. And he was way less good at that position because Kevin is really light, you know, really nine, really striker. And then afterwards, once Abramovich left, he was the nine and he was just incredible. And he actually beat Abramovich record of goals through like best Goliador from, from Paris. And then recently Mbappe took him the title. But, yeah, yeah, that's definitely that.

Ravi Ramineni 01:27:35

Yeah, I.

Alex Andorra 01:27:35

But more generally, I mean, yeah. What you're saying also echoes the topics in baseball. In the end, the name of the game is a lot player projection and also projection from one league to the other. Baseball has a bit more time in a sense, for each player because you don't have to take to put players in the major league as young as football has to do so. You have a bit more time. But, yeah, like the name of the game is the same.

Ravi Ramineni 01:28:12

Yeah. But you're making a decision of who you want to keep in the system to go into aa or single a or three a or something.

Alex Andorra 01:28:20

Yeah. And finally, on a very selfish note, I'm really happy to hear that predictive things, predictive metrics are of very important and interesting because you need models for that. And I'm a modeler and I love modeling. So, you know, I'm like, oh, that's great. That's gonna be a lot of fun and nerdy things in the future coming our way. You heard that, listeners? That's great. Lots of models for sure. Awesome.

Ravi Ramineni 01:28:48

Ravi.

Alex Andorra 01:28:49

Well, you already answered the last question of the show, so the other last question I ask every guest at the end of the show is, if you had unlimited time and resources, which problem would you try to solve?

Ravi Ramineni 01:29:04

I think we just talked about it. Well, first question, is it in football analytics or is it in general?

Alex Andorra 01:29:11

Whatever you want.

Ravi Ramineni 01:29:12

I guess in general, I think I would, you know, I would probably want to solve the problems that we talked earlier about, you know, just people having access to food, water, shelter. I think probably that's probably what I would want to solve if I had infinite resources first, because in the grand scheme of things, sports is a nice thing, but we're not saving lives. We're not sure. We're making some people happy, go home after a game, win and. But I would, I would totally want to change lives of real people in a real way. That would. What I would do.

Alex Andorra 01:29:57

Yeah. Yeah. I mean, definitely, definitely understand that and completely agree. Awesome. Well, Ravi, that was an absolute blast. Thank you so much for taking so much time and indulging all my questions. As usual, I learned a lot. And as usual, we will put resources and a link to your socials, your website, and also the different concepts we've touched on during that episode. Oh, I just that makes me think. While you were talking about priors and talking about coaches and managers about that, I added a link to a python package that's called prelease. That's basically a toolbox for prior elicitation. Definitely encourage listeners and whoever is working on a patient model to use that because that really helps you. Visualizing priors, defining them and the latest release they have done is just absolutely awesome. So definitely take a look to prerelease. It's in the show notes and a lot of other stuff. So thank you again Ravi, for taking the time and being on this show.

Ravi Ramineni 01:31:14

Thanks a lot Alex for having me. It was great talking to you and you know, obviously picked up a few things and I will follow up on with you offline. Yeah, but it's been great. Thanks for your thanks for having me.

Alex Andorra 01:31:27

You bet. Whenever you want. This has been another episode of learning bayesian statistics. Be sure to rate review and follow the show on your favorite podcatcher and visit learnbasstats.com for more resources about todays topics, as well as access to more episodes to help you reach true bayesian state of mind. Thats learnbasstatsats.com. our theme music is good Bayesian by Baba Brinkman, Fit McLasse and Megaran. Check out his awesome work at Baba Brinkmande. I'm your host, Alexandora. You can follow me on Twitter alex Endora like the country, you can support the show and unlock exclusive benefits by visiting patreon.com learnbasedats. Thank you so much for listening and for your support. You're truly a good basie and change your predictions after taking information in and if you not be less than the main, let's adjust those expectations. Let me show you how to be a good daisy. Change calculations after taking fresh data in those predictions that your brain is making.

Ravi Ramineni 01:32:40

Let'S get them on a solid foundation.