Fantastic interview. Thanks for such an insightful conversation into the decision process and thinking behind DUPR algorithm changes. Mad props to Sarah for such a nuanced and thoughtful discussion. Definitely gives me more confidence in the direction the DUPR is heading towards and I'm glad that she's leading the charge!
Great interview, really interesting stuff. It's fantastic to hear Sarah talk about her DUPR passion. I also love the platform, we have our own club and I do my best to promote DUPR as much as possible. I do think it's great to hear that the algoritm will change for the most loyal players. I also do have the feeling I get stuck and that's frustrating. I already have played so many matches, that it seems whatever I do, I will stay stuck with my 4.5 DUPR. Earning wins is making me crawl higher, but really at snail pace. Any idea when this update will be released? It's going to be loved by the most fanatical players.
Thank you for watching and for this thoughtful comment! Sarah is smart, passionate and dedicated to her craft. I believe she said the algo update will be q1 this year, but did not give a specific date. And you are right, it will be loved by the players that care about their DUPR :) aka US :)
Love the term Sarah used about hobby jumping. I am most definitely a hobby jumper... until pickleball. Great interview and thank you for the meaningful content.
OMG I am SO THANKFUL for the rating adjustment today. I have felt like I was banging my head against the wall, getting better, playing in dupr tournaments, good results, feel skill going up, DUPR is like a rock that won't move. All of a sudden today my rating adjusted up .26 to close to where I think my rating should be. We'll see if I really belong there after the tournaments to come. Side note, It will be interesting to see now with these changes, some people that registered for upcoming tournaments in the correct bracket are no longer in the correct bracket based on today's adjustments. But if other brackets are already full, what will TDs do? Let them stay or kick them out and put alternates in that now fit the ratings bracket limits? But my REAL comment/question besides those is....there is another issue that wasn't discussed and I'd really appreciate a response. I feel there is a problem or at least was before the adjustment today that when 2 people on a team are not real closely ranked (even just .15 or .25 apart) in that in a win, the top ranked player gets a LOT more rating point boost than the lower rated player. But then if the same team loses, the bottom rated player goes down more than the top rated player. it seems like the algorithm always rewards/protects the top rated player on a team and punishes the lower rated player. I know there are other factors, like reliability rating, and this was one reason my rating might have moved less points than my partner if my rating had a higher reliability score, but that is absolutely not the only issue involved here. We know who gets most of the balls in a tournament: the weaker/lower rated player. So when I play with a .25 rated higher player and opponents hit all the balls to me b/c they know my teammate is higher rated/better than me, and then we win, and my teammate gets the lion's share of the rating point boost, this is just not right. How does the algorithm account for rank disparity within a team, and how can it improve on this? I was surprised this wasn't asked. But overall very cool interview. My son is an actuary and played college sports and your job sounds way more fun than his, Sarah.
So basically the 1/2024 algorithm update sacrificed accuracy for transparency? I guess I'm the minority that doesn't want that and prefer the old method. There are many rating system already out there in the video game space like StarCraft2(RTS) and League of Legends (MoBA) and Fortnight(Battle Royal) and team shooter games. Some addresses 2v2 rating problem. One of the many things they've implemented is having multiple numbers like your true hidden rating (like a pure elo algorithm) and the rating that is displayed (this is more of a moving average). The rating that is displayed can make sense as in rating goes down when losing and up when winning, but the amount of movement is based on your true rating. The further it is from your true rating the more it moves towards it when you have a match outcome as predicted and the reverse of your display rating moving less when moving away from the true rating. I like the sound of minimum movement, especially when your reliability score is high. My 4.0+ friend with a 3.2 DUPR and high reliability score barely moves when beating 4.0s and losses it all back when they lose to another 4.0. Sounds like with the current algorithm he'll need about 50 matches to obtain a DUPR reflective of his true skill. I also like taking account rally scoring vs traditional. Can't wait for DUPR to roll out these algorithm changes.
Thank you for this thoughtful, and insightful, comment. I had no idea about these nuances in the video gaming space! I would rather accuracy vs transparency as well. The continued algo updates should help. I like that DUPR is willing to change/adjust. Their employees are passionate and smart, like Sarah, so that helps!
Hey! Yeah thanks for the well thought out comment. The ideas you described in the video game space are definitely things we have thought about. When there is a hidden rating, there is of course transparency lost, so that is a con with that method. But it is good to know that we would have at least one user be on board with that type of mechanism :) Yes, we very much value accuracy as well! We have already taken big accuracy strides since that initial change in June 2023, and we will continue making them. Thanks again for the feedback, and hope to keep making DUPR even better for you.
27:50 - DUPR is not fair at all… me and my friend got really close to higher DUPRed players and still our DUPR went down by a lot. I won 3 or 4 games against beginners and my DUPR went way up.
I would like to start seeing the predicted match score along with the actual match score and rating movement data that you see when you click on the match. Also, don't love that a not rated person who plays no one over a 4.25 rating and ends up wirh a beginning rating of 4.6. that makes no sense to me when all of that persons matches were close and not blow outs. How does ones rating get set above anyone else rhey played when they started as NR. Overall love the DUPR system though! And thank you dor rhe most recent update as i was one of those people stuck at 4.0 flat forever.
Maybe I missed it but what efforts are Sarah and the DUPR team doing to improve data acquisition? What efforts are underway to simplify score entries from the pickleball community? For example, tournaments, clubs, and even local play. Thank you
Some feedback I have for DUPR is for many tournaments, a group may sign up at the correct level and one of 2 things happens. #1 is people with much higher DUPR scores sandbag. example: a 3.5 bracket but the gold medal team was 4.2 and 4.4. #2 in smaller tournaments the director sometimes combines brackets because of the numbers involved. Example: a 3.0 team signs up but no one else does so they put them in a 3.5 division. In both cases the team which was signed up to play at the correct level is penalized. I know that is a difficult situation but it seems there should be a way to accommodate that.
We are continually working with club and tournament organizers to help with situations like this as this is really more to do with the way the product is used and less to do with the mechanism of the algorithm itself. For the first example, it is crucial that the director requires players to enter according to their DUPR rating to avoid this. In general, sandbagging can be avoided by directors requiring players play at their DUPR rating and then also adding the matches from those results to move players accordingly. In the second example, this is again on the director for combining the levels. Instead, maybe a format change, like a double round robin, is a more acceptable solution to keep everyone playing at that proper level. Thanks for your feedback and for listening to the podcast!
@@SarahCarpenter-DUPR great podcast! regarding the sandbagging issue, my area recently had a tournament where entrants were policed SOLELY by their DUPR ratings (except skinny singles), and it worked great! A lot of 4.2-4.3 DUPRs I knew were planning on signing up for 3.5 but weren't allowed to. it made the brackets a lot more fair for everyone! Quick question for you, Sarah: any chance of skinny singles tournaments being incorporated into DUPR somehow? I have been manually inputting my skinny singles tournament matches into "singles", but of course people I lose to are validating them, and people I beat are not validating 😂
So, as a long time Chess Tournament player, I have always wondered if the raiting algorithm for US Chess couldn't be leveraged for DUPR ratings. It's been around for a LONG time and works (IMO). Why recreate the wheel? Has Sarah looked at existing rating algorithms? Has she ever looked at what the US Chess federation (organization) uses?
@@aaronsoto1504 Hi! Actually it does reflect the complexity of the point spread (four digits). To borrow from their algorithm you would either combine ratings of players (3.54 + 4.01 = 7.55) or average them. And currently DUPR is only a win/loss algorithm since you don't have ties - and per the video, it's not reflecting the complexity of the point spread.
Is this going to fix things like my friend who played one tournament with a high level partner and now has a 4.0 rating that is NOT reflective of their skill?
I would rather let DUPR speak but from what I know, the more data DUPR has on the person, the more accurate the rating. So although this friend played one tournament and it shows a 4.0 rating, her reliability score should be very low.
Yes, Kaitlyn is spot on. With only one tournament, they will have a low reliability score, and it will decrease over time as long as no new data comes in. If they play their next event at her 4.0 rating and they can hang, they will remain at that level. But if they lose at this level, the rating will correct itself and lower. The key is playing at your rating level and then adding the matches so that the rating can properly adjust. Then rinse and repeat with the new rating :)
Fantastic interview. Thanks for such an insightful conversation into the decision process and thinking behind DUPR algorithm changes.
Mad props to Sarah for such a nuanced and thoughtful discussion. Definitely gives me more confidence in the direction the DUPR is heading towards and I'm glad that she's leading the charge!
Great interview. Thank you to both of you for your time and thoughtfulness.
Great interview, really interesting stuff. It's fantastic to hear Sarah talk about her DUPR passion. I also love the platform, we have our own club and I do my best to promote DUPR as much as possible. I do think it's great to hear that the algoritm will change for the most loyal players. I also do have the feeling I get stuck and that's frustrating.
I already have played so many matches, that it seems whatever I do, I will stay stuck with my 4.5 DUPR. Earning wins is making me crawl higher, but really at snail pace. Any idea when this update will be released? It's going to be loved by the most fanatical players.
Thank you for watching and for this thoughtful comment! Sarah is smart, passionate and dedicated to her craft. I believe she said the algo update will be q1 this year, but did not give a specific date. And you are right, it will be loved by the players that care about their DUPR :) aka US :)
Thank you for the kind words and for watching! Yes, the update is coming soon soon :)
Love the term Sarah used about hobby jumping. I am most definitely a hobby jumper... until pickleball. Great interview and thank you for the meaningful content.
Thank you for your kind words, Kevin! Keep pickling :)
OMG I am SO THANKFUL for the rating adjustment today. I have felt like I was banging my head against the wall, getting better, playing in dupr tournaments, good results, feel skill going up, DUPR is like a rock that won't move. All of a sudden today my rating adjusted up .26 to close to where I think my rating should be. We'll see if I really belong there after the tournaments to come. Side note, It will be interesting to see now with these changes, some people that registered for upcoming tournaments in the correct bracket are no longer in the correct bracket based on today's adjustments. But if other brackets are already full, what will TDs do? Let them stay or kick them out and put alternates in that now fit the ratings bracket limits? But my REAL comment/question besides those is....there is another issue that wasn't discussed and I'd really appreciate a response. I feel there is a problem or at least was before the adjustment today that when 2 people on a team are not real closely ranked (even just .15 or .25 apart) in that in a win, the top ranked player gets a LOT more rating point boost than the lower rated player. But then if the same team loses, the bottom rated player goes down more than the top rated player. it seems like the algorithm always rewards/protects the top rated player on a team and punishes the lower rated player. I know there are other factors, like reliability rating, and this was one reason my rating might have moved less points than my partner if my rating had a higher reliability score, but that is absolutely not the only issue involved here. We know who gets most of the balls in a tournament: the weaker/lower rated player. So when I play with a .25 rated higher player and opponents hit all the balls to me b/c they know my teammate is higher rated/better than me, and then we win, and my teammate gets the lion's share of the rating point boost, this is just not right. How does the algorithm account for rank disparity within a team, and how can it improve on this? I was surprised this wasn't asked. But overall very cool interview. My son is an actuary and played college sports and your job sounds way more fun than his, Sarah.
once again fabulous podcast! Sarah is amazing! So great to hear someone in pb who says we can always improve and she is doing it!
Thank you so much for watching and supporting this podcast. I really appreciate it!!
Thank you for watching and for your kind words. Glad that pickleball has encouraging people like you :)
Very interesting. Good idea for an interview.
Thank you!! Sarah is fantastic
Great information! Thanks for creating this episode.
Shoutout to my fellow OG UNC Club Member!!! Let’s go Sarah!
So basically the 1/2024 algorithm update sacrificed accuracy for transparency? I guess I'm the minority that doesn't want that and prefer the old method. There are many rating system already out there in the video game space like StarCraft2(RTS) and League of Legends (MoBA) and Fortnight(Battle Royal) and team shooter games. Some addresses 2v2 rating problem. One of the many things they've implemented is having multiple numbers like your true hidden rating (like a pure elo algorithm) and the rating that is displayed (this is more of a moving average). The rating that is displayed can make sense as in rating goes down when losing and up when winning, but the amount of movement is based on your true rating. The further it is from your true rating the more it moves towards it when you have a match outcome as predicted and the reverse of your display rating moving less when moving away from the true rating.
I like the sound of minimum movement, especially when your reliability score is high. My 4.0+ friend with a 3.2 DUPR and high reliability score barely moves when beating 4.0s and losses it all back when they lose to another 4.0. Sounds like with the current algorithm he'll need about 50 matches to obtain a DUPR reflective of his true skill.
I also like taking account rally scoring vs traditional.
Can't wait for DUPR to roll out these algorithm changes.
Thank you for this thoughtful, and insightful, comment. I had no idea about these nuances in the video gaming space! I would rather accuracy vs transparency as well. The continued algo updates should help. I like that DUPR is willing to change/adjust. Their employees are passionate and smart, like Sarah, so that helps!
Hey! Yeah thanks for the well thought out comment. The ideas you described in the video game space are definitely things we have thought about. When there is a hidden rating, there is of course transparency lost, so that is a con with that method. But it is good to know that we would have at least one user be on board with that type of mechanism :)
Yes, we very much value accuracy as well! We have already taken big accuracy strides since that initial change in June 2023, and we will continue making them. Thanks again for the feedback, and hope to keep making DUPR even better for you.
@@SarahCarpenter-DUPR Hi Sarah, thanks for your hard work! And thank you both for this episode.
27:50 - DUPR is not fair at all… me and my friend got really close to higher DUPRed players and still our DUPR went down by a lot. I won 3 or 4 games against beginners and my DUPR went way up.
Play more games so that the algorithm has more data on you. It won't swing so much once it has a higher confidence on your rating.
I would like to start seeing the predicted match score along with the actual match score and rating movement data that you see when you click on the match.
Also, don't love that a not rated person who plays no one over a 4.25 rating and ends up wirh a beginning rating of 4.6. that makes no sense to me when all of that persons matches were close and not blow outs. How does ones rating get set above anyone else rhey played when they started as NR.
Overall love the DUPR system though! And thank you dor rhe most recent update as i was one of those people stuck at 4.0 flat forever.
Maybe I missed it but what efforts are Sarah and the DUPR team doing to improve data acquisition? What efforts are underway to simplify score entries from the pickleball community? For example, tournaments, clubs, and even local play. Thank you
Some feedback I have for DUPR is for many tournaments, a group may sign up at the correct level and one of 2 things happens.
#1 is people with much higher DUPR scores sandbag. example: a 3.5 bracket but the gold medal team was 4.2 and 4.4.
#2 in smaller tournaments the director sometimes combines brackets because of the numbers involved. Example: a 3.0 team signs up but no one else does so they put them in a 3.5 division.
In both cases the team which was signed up to play at the correct level is penalized.
I know that is a difficult situation but it seems there should be a way to accommodate that.
We are continually working with club and tournament organizers to help with situations like this as this is really more to do with the way the product is used and less to do with the mechanism of the algorithm itself. For the first example, it is crucial that the director requires players to enter according to their DUPR rating to avoid this.
In general, sandbagging can be avoided by directors requiring players play at their DUPR rating and then also adding the matches from those results to move players accordingly.
In the second example, this is again on the director for combining the levels. Instead, maybe a format change, like a double round robin, is a more acceptable solution to keep everyone playing at that proper level.
Thanks for your feedback and for listening to the podcast!
@@SarahCarpenter-DUPR great podcast!
regarding the sandbagging issue, my area recently had a tournament where entrants were policed SOLELY by their DUPR ratings (except skinny singles), and it worked great! A lot of 4.2-4.3 DUPRs I knew were planning on signing up for 3.5 but weren't allowed to. it made the brackets a lot more fair for everyone!
Quick question for you, Sarah: any chance of skinny singles tournaments being incorporated into DUPR somehow? I have been manually inputting my skinny singles tournament matches into "singles", but of course people I lose to are validating them, and people I beat are not validating 😂
I wanna see the match expectations be shared.
So, as a long time Chess Tournament player, I have always wondered if the raiting algorithm for US Chess couldn't be leveraged for DUPR ratings. It's been around for a LONG time and works (IMO). Why recreate the wheel? Has Sarah looked at existing rating algorithms? Has she ever looked at what the US Chess federation (organization) uses?
Chess is a win/loss/tie algorithm which does not reflect the additional complexity of the point spread.
@@aaronsoto1504 Hi! Actually it does reflect the complexity of the point spread (four digits). To borrow from their algorithm you would either combine ratings of players (3.54 + 4.01 = 7.55) or average them. And currently DUPR is only a win/loss algorithm since you don't have ties - and per the video, it's not reflecting the complexity of the point spread.
DUPR SARAH!!!!
She's THE best!!
Is this going to fix things like my friend who played one tournament with a high level partner and now has a 4.0 rating that is NOT reflective of their skill?
I would rather let DUPR speak but from what I know, the more data DUPR has on the person, the more accurate the rating. So although this friend played one tournament and it shows a 4.0 rating, her reliability score should be very low.
I agree with you. In this case it works against the if they want to compete against their level competition.
Yes, Kaitlyn is spot on. With only one tournament, they will have a low reliability score, and it will decrease over time as long as no new data comes in. If they play their next event at her 4.0 rating and they can hang, they will remain at that level. But if they lose at this level, the rating will correct itself and lower. The key is playing at your rating level and then adding the matches so that the rating can properly adjust. Then rinse and repeat with the new rating :)
Sarah is the GOAT 🐐🏓
in a lot of games, there's elo and hidden elo
Talk DUPR to me
Not as deep as I would have gone. But I might be nerdier than Sarah.
A girl being in charge of the algorithm explains so much
Your comment about a girl being in charge of the algorithm explains so little.