From Rules to Reality: Python's Role in Shaping Roundnet

Speaker 1 [00:07]

I'm really honored and I have a really nice and chill topic for you which will quickly go really deep into sports and sports analytics with Python so I hope you have fun and yeah enjoy some also some nice video snippets from my sport which is weird but nice to watch so about myself you maybe know me I'm I'm Larissa. I'm also part of the organization team of PyCon.de. Usually I'm a data scientist at Savanta AG and also a team lead for my automation team. And I studied political science and data science and I like discussing about AI a lot. So what I do in my free time, I get myself involved in more data stuff, but in sports. So this is also me. This is me in my round net life, which takes most of my free time, where I have been playing round net for four years now in Mannheim, at the Mosquitoes Mannheim. And you can see here, like, round net is a very popular sport at universities. You might see it in the park. I will come to that later. But we also have a really nice Django website, where we organize all the tournaments, and we have some nice game statistics. and the ELO rankings, so everything in Python, just a side note. This talk will not be about that ELO ranking, sadly, but about our rule change. Because what is round net? You might have seen round net. Round net is a cool sport you can play in the park. You have a small net on the ground and a ball that's just like this size, and it's played like beach volleyball. So you have three contacts per team, and you try to hit the ball on the net instead of over the net. This rally actually is the most viral rally on Instagram last year. So this is a very good example how round that could look like. And it's very engaging and spectacular. People are flying around and running like crazy. And it's a lot of fun. If you have seen it maybe in the park, you might also known it as Spikeball. Spikeball is a big manufacturer of that sport. And if you take one thing home, also those on the video right now, if you take one thing home, don't call it spike ball, please call it round net. It's the official term of the sport. And there are also other manufacturers like Premier Spike with the blue logo here and this is actually also a Premier Spike set that you can see here. Okay, but let's go back to my talk. So I will be talking about the current problem that RoundNet has because there is a problem and how we are going to solve this and what Python has to do with this all. So stay tuned. So what is the problem? You can see here the serve isolated by a lot of times and you can see here that there are a lot of aces. Like people are hitting the serve on the net and it gets a spin on the ball and the spin is causing a change of direction so that the other player cannot reach it. So this is regarded as an ace. An ace is an automatic point for the serving team and it's very hard to defend. Like if you cannot touch the ball, you cannot return the ball to the net by yourself and make a point by yourself. So an ace is very powerful and unfortunately it's very strong and very common because people are just grinding serves and drilling like this special spin on the ball and it's a big problem. And it's also not very enjoyable to watch because four people are standing around the net and nothing really happens. another part of the problem is what you can see here the surface happening three touches and then the hit happens but you don't have a rally it's just a side out it's a clean side out where the receiving team makes a point but there is no defensive touch there is no action there is no blocking or no like crazy touching because there's just a hit on the net and the point is done So we want to increase the rally size here so that we have less aces and less side outs. And the third part of the problem is because aces are so strong that people are double faulting a lot. You have two times the chance to hit the ball correctly on the net. And if your serve is not correct, you can have a second chance. And so you have, when you try to really ace a lot, you also have a very high double fault rate. And a high double fault rate leads into games like this. This is actually the women's final of US Nationals last year. And they play 10 points without a legal point interaction. Like it's just double faulting. 10 points. They're just standing around the net and nothing happens. And it's not very enjoyable to watch. And we see this trend increasing over the last years because people are getting more experienced with serving and with the ball and they get really specialized. So we try to change the rules that it gets more fun to watch. So what happens now? So, basically, the whole RoundNet world decided, okay, we need to change something. So, you can see here some of the national government bodies, like RoundNet Brazil, RoundNet Hong Kong, Luxembourg, RoundNet Germany, like all of them are in this group. You also have the European RoundNet Association and the manufacturers, and they are all part of the RoundNet rules revolution. So we try to really have a revolution this year and change the round net rules. Because we want to make round net more engaging, you can see here a really nice example how round net could look like. We want to increase the difficulty for attackers so we have more defense touches, that the defense has a better chance. We want to have more rallies, more crazy touches, more run-down balls. And we want to take the focus of the game away from the serve and back to the rally itself. And the crazy thing is, we are doing this with data. Imagine you would do this in soccer or in football, just measuring game statistics, view statistics, and just say, okay, hey, we noticed on a smaller field, viewing statistics are better. So we change it like we don't care. We just change the rules So what is actually our approach and here it gets like more data involved and you will also see some Python code at the end We created a baseline so we watched Games and annotated them and counted points and double faults and aces and defensive touches is to have a baseline that we can compare the new rules to. Then we test the rules in different tournaments all over the world, like we have test tournaments in the US, in Japan, in Brazil, in Mallorca, in Germany, like all over the world. And then we compare the results and decide which rules we want to roll out next year globally. I know this sounds crazy and I feel your excitement about that so let's dive deeper into how our test setup looks like so we have two test packages that are currently played so if you would would play a round test round a tournament right now in Germany it was most probably be a test package for package a and then we review the first results we adjust the rule set in the last package and then we decide and during that we have like a constant stream of data annotations and of data analysis to keep track of all the changes that we have with the new rules and this is our data pipeline as we are volunteers all of us and we do this now free time and we have no money, our database is Google Sheets. We collect the data there from the baseline that we annotated. We get the data there from videos that were uploaded by players after test tournaments, so we annotate them as well in Google Sheets. And then we have two Google surveys where we collect also player feedback, because we also want to measure is it actually fun to play with the new rules, or is it stupid and the new rules are just annoying? And then we do some data analysis on it, on very nice Google Collab notebooks right now, maybe this will change, to compare the results at the end. And this is how it looks like. It's nice, isn't it? So we have the first row is the change of possessions, this is very technical, I'm sorry, change of possessions, how often the ball passes the net in a legal way. Then we have the last possession touches, like how many touches the last team had on the ball, like without returning to the net, so we can also count defensive touches that were not successful. And then we have a double fault checker, so we can also check annotate for double faults. And we are also counting if it was a redo. So for example, if there is an error but the other team does not agree that it was an error, there is a redo because there is no referee to say what was the reality. So the point is just repeated, and we also keep track of that here. Okay. So you might want to look at first results. And it's really like it's really crazy. We analyzed in the baseline 73 games in total. This is over 17 hours of gameplay that we just watched and annotated, and this is only because we had condensed videos, which were not full length, because otherwise it would have been crazy. So this is the baseline you can see here. We have three categories, which I need to explain. We have coed, which is one man and one woman is playing together as a team. We have the open category, where basically everyone can play, and we have the women category, which is only for women. And you can see here already the differences in the values that we measured. So for example, aces are very high in coed. And this is a problem because we still have a difference in serving skill and in receiving skill between men and women's games. So when we play coed where one man and one woman is playing together at a net, you have for the women, very high difficulty to receive the serve because it's very hard to get to it because it's very technical. So here the ace value is very high. And we also have a lot of double falls, especially for the men or for the open category here because they're just really trying to push for the ace because it's so much worth when And you can just hit an ace in comparison to just put the ball on the net and it's easy to reach for the opponent. And you can see here the small area which are the rallies. I mean, look at this. And this is the percentage size. Like we have less than 10% rallies in the game. So every tenth point is a rally. This is not enough. We want to have it like every point should be a rally. Every point should be spectacular. And if you add up the double falls and the aces, you can see, like, nearly 50% of all the points are just people standing around the net and one person is hitting either with an arrow or with an ace. And then you change possessions and then it's done again and you still just stand around the net and nothing happens. So this is not very nice to watch. And I can tell you, it's also not very nice to play, except you just like to grind serves on your own, and you're really good at it. And for comparison, I put here some stats from other sports, because it might be difficult to compare it. So for tennis, for example, we have a double fault rate below 10%, which is, it differs for men and women, but this is like, I think, men at the highest level. And for beach volleyball, which I think is the most similar game to compare Round 2, it is around 13, 15% like a surf fault rate, so we are still below our rate here. As well for the aces, you can see here in the middle, it's far more seldom in other sports to have an ace than in round net. One thing that I found particularly interesting was the comparison of clean side outs. Because if you ever watched a beach volleyball game and it contains nearly 40 to 60 percent of clean side-outs. You can kind of grasp that it is not very spectacular when you have a serve and then a clean side-out because it's just, yeah, another point and another point. And it's also a criticism of beach volleyball that they have not enough rallies, that they have not enough interactions because it's just a side-out game. So we see here they have similar problems, and we are actually doing something about I already talked about that. We now have the baseline. What do we want to achieve? We want to include all three kinds of involved persons in that whole rounded tournament sphere. So we want to increase the player satisfaction that they have fun while playing. We want to have the spectators have a nice time because they like to watch the sport because we also want to see round nets sometimes maybe on TV or at the Olympic games. So it would be very cool that the spectacular value is high of that game. But we also want to keep it easy for organizers to set it up because round net is a very community driven sport and all the tournaments are organized by volunteers, by community members. So if we had like very complex equipment like a camera to measure, I don't know, some errors, this would be hard to set up and this would be not feasible by the local communities. So we came up with some measurements that we actually can maybe grasp a little bit better. So here you can see the touch per point relationship, like how many touches does the ball have per point on average for the single divisions that you can see here. And if you look at the scale below, you can see that the average of open is under three. If you imagine, a clean side out would be one serve, which is one touch, and then three touches to bring the ball back on the net, so four touches. So we are on average even under a clean side out rate. So we are on average under four touches. And this is just, it's just nothing. For women, it's a little bit better. So we are here over three, but as long as we're under four, we're still having more aces and more double falls than actual rallies. And this is not the ratio that we want to have. So now we actually had also our first results from package A because the tournaments are just rolling out and the results are just coming in and this is actually a screenshot from yesterday where I included the last data that we got here and I haven't brought it in a nice visual form because. And you can see here already some results. So package A had some really nice rule changes included. For example, we included an out of bounds zone. So usually round net is played without any boundary. So you can play like until infinite, basically. And there is no out of bounds. But here with package A, we included this out of bounds to see if it could boost defense because they were not just over boomed by some balls, because they were just then out of bounds. What we can see here, in comparison to the average of the baseline, the double faults actually increased, the ACEs actually increased, the clean side out rate was lower, so this is a little bit my mistake, but so for the ACEs we don't have a result. You can also see that the standard deviation is higher because the package A results are still small and they are also very small for the single division, so I cannot do a breakdown into divisions right now because we just have not enough data. But what we can see, alone this small rule change with this package A already increased the number of touches by a lot on average. It also increased the rally number and the rest like related to that touches per rallies or touches per point ratio also was increased. So we see it goes into the right direction kind of but we're not finished because I mean package B will be very nice and we're waiting for the results there in package B. And I mean, the real work starts now, because now the videos are dropping into our Dropbox and we need to watch them and we need to annotate it. So if any one of you has some free time. But we are ten very motivated RoundNet players that actually happen to be data scientists as well. I don't know if this is a coincidence, but we have a lot of IT people in RoundNet. And we are very keen to just do a nice data analysis to keep this rule package sorted. We have 22 test events scheduled right now. And we already have a lot of community responses. That number actually is not correct. I'm sorry. Because people are so involved into this topic, because they love their sport so much, that They really want to contribute and they want to get this sport a better sport for them to play and to watch. And I think this already gets me to the end. I was a little bit faster maybe to get you earlier into the weekend, but maybe we can have some discussions and some nice questions later. So if you have any questions, let me know or contact me. Thank you.

Speaker 2 [21:49]

Yeah, so thank you for this learning between Spikeball and Netball. So, yeah, the first question from you is, did you have time to check for statistical significance of the package A rules based on a small sample size?

Speaker 1 [22:07]

No, like the number of videos we got is in any case not enough to say anything statistically relevant right now because it's just too small. Also, this whole statistical setup is a mess. I know because we cannot control weather, we cannot control other conditions. We have so many factors that we cannot control. so my statistical heart is a little bit broken but this is reality and so we try to do as best as we can but the package a in any case has none not not a statistical significance right now because it's just not comparable right now

Speaker 2 [22:52]

Okay, the next question is, do you also get player feedback where they... It's removed, sorry.

Speaker 1 [22:59]

Sorry.

Speaker 2 [23:02]

It's disappearing currently It's back No, okay Do you also get player feedback where they themselves propose rules rule changes?

Speaker 1 [23:18]

So this was actually something that started last year already. We had several surveys beyond the round net players all over the world where we first asked what are pillars in round net that you like and that we should not change with a rule change. Because there are certain things that are just nice about the sport. Like the 360 degree angle that you have. This is something that we don't want to change. We don't want to have, like, separate sides or something like that. Or, for example, to have this very... The focus on agility, we don't want to smurf because this is something that makes this sport very fun. So this was the first survey. And the second one was a rule... Yeah, we asked for possible rule changes. And I don't know, I wrote it here. we had in the survey around 600 replies that said hey you could change this or you could change that or I tried out this and then there is also the third thing that is happening because there you cannot change just the rules but you can also change the equipment like we're discussing a bigger ball because bigger balls have less velocity, you can put less spin on it, so it gets easier to reach it, the touches get better. So this is also something that we discussed and that players proposed because they just tried the game out with different balls. So yeah, community is very included here. And all the rule changes in the packages are based on community proposals, basically.

Speaker 2 [25:04]

Thank you, so next question is what are all the rules rule changes for B?

Speaker 1 [25:11]

Okay now you're testing me. So we have a lot of rule changes. For B we have again the out of bounds zone. We have a table tennis serve. So right now it is like in beach volleyball when you serve and you make the point, you serve again. And for package B, we will have table tennis serves. So you have two serves in a row and then the serve switches to the opponent. And then they have two serves in a row. We have different defense movement patterns that we now allow. And we have a bigger no-hit zone. Okay, maybe I can explain the no-hit zone because this is really interesting. So, if you imagine this blue dot in the middle is the net and you want to hit the ball. If the ball is perfectly in the middle, you have 360 degrees to hit the ball, which makes it very hard for the defence. If you have the ball here and you hit it, you immediately have less angle in the hit, So the defense has a better time and has a higher chance to reach the ball. What we introduced some years ago was a no-hit zone, a zone around the net where you're not allowed to step in when you hit, which makes it more difficult for the attacker to hit. And we want to increase that zone, so it's really hard to hit a ball for this person when the ball is here, because you have to basically fly or jump over the net to hit it, which are people actually doing. So those are some of the rules that we have in package B.

Speaker 2 [27:11]

So, next question. Have you considered annotating clips with VLMs?

Speaker 1 [27:17]

Yeah, I get that question a lot. The thing is, if someone has some resource to do this AI thing on the annotation task, please let me know. We just thought it's cheaper to use humans to annotate it. But yeah, it's still an option. At least after this project, we have a lot of training data with annotated videos. So there are actually also data science people who are training models, especially for round net game analysis, but we don't have access to that. So if someone has, let me know.

Speaker 2 [28:02]

So, next question is, rallies are probably more exhausting for players than aces. Do you think this will require the game to change in an additional way?

Speaker 1 [28:14]

Yes, yes it will. We see through the last years that the game gets more and more athletic. Like you have to be, when you want to play in a high division, you have to be really fast, you have to have really fast feet and fast legs to run, to sprint, to change directions, to react. and this will get more and more important like you I just said at some points you have to jump over the net to hit the ball and this will be more and more important so also the endurance like over a long time when you don't have that much aces yeah this will be this will change the sport but this is our goal we want we really want to change the sport to make it into something really nice and enjoyable

Speaker 2 [29:05]

okay so maybe last question in your technology background you had a chat DVD icon how is that used in your analysis

Speaker 1 [29:16]

So we have a lot of open questions in our feedback forum because this is something where we really want to focus on to integrate community feedback into our analysis. But it's really hard to go through hundreds of open text feedback in like 30 languages. So we get feedback from France, from Japan, from Brazil, from Russia, and it's like this is not working out without AI. So we are using an LLM to first translate the open text answers and then as well to summarize them. So we actually know what they wrote in the survey.

Speaker 2 [30:10]

Thank you. So we are over, but yeah, it's like the last question. So take it with. So the last question is, did you consider limiting the area that has to be struck for serve, meaning where on defense is allowed to land? Sorry, I'm just a little bit.

Speaker 1 [30:27]

Yeah and no. So we are limiting the area that the server is allowed to step while serving. So I heard that in South America, there's actually the pattern that people are starting like this, and you're allowed to do one step while serving, so they do it like this. And then they serve, which leads, with the spin on the ball, that they have nearly, they shoot their own player. They start here and they serve and they hit their opponent because they have so much angle. So we are trying to limit the step size or not allow a step at all or not allow the split step. So these are all variations that we have in our rule package somewhere in the future. Okay. Thank you.

Speaker 2 [31:22]

Thank you. So then, again, applause for this great presentation.