Autocategorization Test Events & FAQ

Kind of, if you have a race with 100 people in it, and a race with 20 in it, the front group\podium probably will be around the same time, its when you move out of these groups you will see discrepancy as the race with larger numbers will have more ‘blobs’ forming therefore a quicker race time. Plus add in primes or points races and this then can also affect the race dynamic and therefore time to complete.

Time to complete a course is only relevant to that specific race when looking at bunch or scratch races.

Yes, only talking about bunch or scratch races, not individual efforts. That would be an apples to oranges comparison.

Performance is the ultimate predictor of performance.

Elo systems (like ZP’s already existing Race Ranking and MS TrueSkill) also compare you against riders who were not in the same race because there is no guarantee that the stronger rider has ever ridden the course faster than the weaker one. These are solved problems.

That is why many of us are trying to puzzle out what the point of the system in testing is.

5 Likes

Yeah, except we don’t know how someone performs in a race until they perform, do we? Why don’t we all stop pretending like there is one true way to predict someone’s performance in a race, however that may be defined (results, time, power, etc.)? The name of the game is data. Whichever gives the least error in a data fit will be the best predictor and we don’t know what that is until someone actually does that analytics exercise and shares the result. My understanding is that this whole testing thing is basically that data exercise: Try different data points to predict people’s speed around the course, race results, etc. to see which model would yield the best categories.

1 Like

Exactly what Geralt mentioned. The name of the game is data.

I think the work that WTRL is doing together with Zwift is a great initiative for the evolution and improvement of racing in Zwift. I participated in the two Classics Races, the first one I found the results of autocat fair but the second uncalibrated.

I believe that the winning prediction algorithm which is being used in ZRace can potentially provide more accurate results as more factors are calculated (ranking, winning history, w/kg, etc).

I also believe that with a few small tweaks to this algorithm we could have a definitive mechanism for auto-categorization and potentially an Elo System.

For those who do not know ZRace, this is the link to my article on the development and statistical models used to predict winners in Zwfit races. Link: https:// bit.ly/3eOV9hL

I make myself available to WTRL to help.

3 Likes

Brilliant work! 95% accuracy for top 5 finishes is admirable precision. The problem WTRL is trying to solve is a bit different though. Your tool predicts race results in each category, taking assigned categories as given. WTRL, on the other hand, is trying to place people in categories that would yield the best distribution of skill. So, in their case, category is endogenous to the solution. I still think WTRL can benefit from your experience in building a model to predict race results.

Geralt - In fact, almost everything is ready. Today the algorithm predicts by category because that was the purpose for the ZRace tool.

To explain in a simple way, today the algorithm can predict a race regardless of the category. With this prediction, it is possible to see riders from low categories in high positions and vice-versa. Then, as I mentioned, with few tweaks and logic rules to distribute the new categories, the mechanism will be complete and ready to test.

I think stopping to pretend there’s a way to (accurately) predict would be a good start, yes. (Fully accurate prediction isn’t even a desirable feature, were it possible.)

That’s what Ranking is a about.

Results are the only data. There’s literally nothing else needed. If you want a coarse mechanism to seed cats initially for those without a race history, that’s fine, and it’ll self-correct.

Coming up with ever-increasingly complex system on top of that is, on the face of it, unnecessary. Or guesses at what it might be, to be more accurate.

I’m baffled by the defensiveness in general. It’s great that something is being done, but that’s no reason to not look critically at the approach.

4 Likes

Is the plan to split female & male riders racing long term?
Otherwise what benefit do you get out these tests when you are undertaking the tests in an environment that wont be replicated in the normal day to day usage?

This pushes the boat 10 weeks or more down the line and doesnt seem to be a long term solution, in addition to making the high profile classics race a bit of damp squib for (some) females who would normally race in the mixed cats but now race with a group of 5-10 other riders rather than a competitive large group.

Still the lack of communication is deafening but hey ho this is the improved forum feedback mechanisms…

1 Like

Does it predict the finishing times then?

They are not. ZP collects a ton of data on power metrics, weight, height, age, course time, and heart rate. I think you are confusing data with outcomes. Yes, result is the only outcome, duh!

There is no need to predict performance, apart from roughly to initially seed someone. After that, a rankings systems looks after itself.

4 Likes

Hey everyone,

Dropping in before the weekend to let you know that we’re still looking at all the feedback here, and we appreciate the ongoing discussion. Your questions and feedback aren’t falling on deaf ears, we’re just working out the best way to get out some responses to you all. That’ll be happening soon!

For the moment, I have two points for you all to specifically bring to your attention:

  1. For the women racing: Would you like to race among the men using this autocat system, or are the women’s categories sufficient enough for you to have a good racing experience?

  2. We have a feedback form we’d love for all of you to fill out should you wish to now that you’ve got a couple races under your belt with this system.

We love to see you all racing and having fun, as well as seeing your questions and discussion. Thank you for helping out!

For those of you who race, we salute you.

4 Likes

As an IRL race organizer, I’ll comment on women’s participation at our events (grass-root road and crits, about 30 per year.)

Most women wanted their own events.

But a small cadre of high end (pro, semi-pro, Olympic level, etc) preferred to race in the men’s events.

To solve this we simply have two categories (at each level), Open and Women. So Cat 1/2 Open and Cat 1/2 Women, Cat 3 Open and Cat 3 Women.

This mostly kept everyone happy.

3 Likes

Feedback from Socks4Watts - it really depends on the numbers. The ladies categories worked well for ZRL where you knew the fields were going to be significant, but the ladies races were practically empty for the classics, and with the autocat they felt like that should seed them properly anyway so would prefer it is mixed gender. With a rankings system you could filter by gender, so are racing for overall points but can see your rank by gender/age etc. Ultimately choice is the best answer - determined by the race organiser. For the Classics combined genders would have been more suitable.

2 Likes

For one thing, I haven’t seen any sandbagging to speak of in women-only races. (Of course I myself tend to do the opposite and race above my assigned category most of the time anyway so what do I know…)

In many women’s races the fields are so small it’s almost so that categorization in general is something of a double-edged sword, if the fields per category are not big enough to create much of a race, especially if each category races alone. Based on what I have heard (haven’t done any myself), this has been a problem with many of the women’s autocat races as well. Not saying there shouldn’t be any categories at all, but the problems to be solved are different.

Speaking of problems to be solved, part of this overall discussion sounds to me like some people just want a category system where nobody (or as a minimum not they themself) ever finishes last in any category.

5 Likes

Before I got an Apple Watch I tried a Garmin VivoSmart 4. Mostly for O2 stats overnight.

But since it was on and recording and pushing to GarminExpress etc, I ended up comparing a Zwift session heart rate as recorded by the VivoSmart 4 to the Garmin chest strap. I simply was not recording much above 130 and for a Zwift race I average about 155 with a range of 135 to 170. If it is not recording anything (nothing!!!) over 130 for that 30-40 minutes, it is simply broken.

That made me think the (newish) VivoSmart 4 was broken so I got it exchanged. The replacement turned out to be the same, the second replacement after that ditto.

About that time I got an Apple Watch. Much better by comparison to the VivoSmart 4, but still not quite as accurate as the good old chest strap. But would (for this discussion) probably be good enough.

VivoSmart 4 is hanging out on the desk. Can’t see that it is fit for purpose.

Some in our team specifically prefer mixed racing. Others would normally race women’s only but have been put off the Classics series so far due to low numbers in the women’s category.

I think the option to choose mixed or women’s only racing would be great. So far, the women’s fields in the Classics have been very small. Of the women I’ve spoken to, we know we’re in the right category, but there simply hasn’t been enough people to form the different groups in the race. Lots of us have ended up riding on our own which isn’t much fun.
The option to join a mixed race with a bigger field would be great.

2 Likes

Wouldn’t this just result in even less riders in the woman-only categories?