What the Data Says About City Streets Small

RexKwonDoRexKwonDo Posts: 63 ✭✭✭
edited February 2019 in General Discussion
Thanks for the warm welcome to this forum. As there was some interest in the data I collect, here is my first official post outside of the IG forum.

City Streets Small is the last of the courses I've analyzed from Japan Atsugi Track 6. It's a very common course in Top Drives, and we've all cringed at one time or another when this course pops up and our hand is full of low ground clearance cars. The question I've had is whether I could find a simple way to know whether my low clearance car can compete with a medium or high clearance car.  

So, I recorded course times for 88 low clearance cars, 79 medium clearance cars, and 58 high clearance cars and began analyzing this data.  It's no surprise that low clearance cars get hung up on speed bumps so they will have one mode of behavior.  I thought about treating medium and high clearance as the same group, but a test for equal variances and a 2-sample T-test showed that each group behaves differently.  It may be due to so many high clearance cars being in the low 0-60 and low grip ranges. Regardless, the data suggested I should treat high and medium as different.  The result is that on this course there are three models to predict course times, one each for low, medium, and high ground clearance.

For each level of ground clearance I ran the data through regression analysis to determine which of the main stats found on the cards are significant predictors of the results.  With that information I then built a useful equation for each of the three ground clearances to predict course times. The equations (found at the bottom of this post) are a bit complex but they are very accurate for predicting. It is natural to have variation in actual times vs. the model since there is information about each car that we don't see on the cards. However, 97% of my actual times are within 0.5 seconds of the predicted times.  72% are within 0.2 seconds. 

But how to make any use of this?  There are too many factors to visualize the full equations in a simple way.  But at all three ground clearance levels 0-60 was a critical predictor. So, I found a different equation based solely on 0-60 time for each clearance height. It's a less accurate equation, but I it becomes useful when I stack all three lines on the same chart so you can eyeball a rough break-even time based on 0-60.  For example, your LOW clearance car with a 3.0 second 0-60 time should be competitive on this course with a MED clearance car with a 5.6 sec 0-60 time.  Grip will add variation, but this chart gets you in the same ballpark. 


For any who want to see my full regression equations, here they are:
  • City Small, LOW Ground Clearance (R-sq=99.31%): 47.105 + (2.022*0-60) - (0.12534*Grip) - (0.00259*Weight)  - (0.04804*(0-60)^2) + (0.00000067777*Weight^2) + (0.000209*0-60*Weight)
  • City Small, MED Ground Clearance (R-sq=99.34%): 50.27 + (1.263*0-60) - (0.2956*Grip) + (0.687*Width) - (0.00435*PeakPower) - (0.01993*(0-60)^2) + (0.000985*(Grip)^2) + (0.000515*0-60*PeakPower)
  • City Small, HIGH Ground Clearance (R-sq=99.79%): 48.50 + (1.7303*0-60) - (0.2967*Grip) + (0.001222*PeakPower) - (0.03851*(0-60)^2) +(0.001039*(Grip)^2)
Anyway, I hope some of you will find this helpful.
Post edited by RexKwonDo on
«1

Comments

  • Nacho101Nacho101 Posts: 535 ✭✭✭✭
    Nice thanks, interesting to see how much of an impact 0-60 time has over grip. Im guessing the time must still vary considerably when comparing cars with same 0-60 but huge difference in grip/weight - that would fall outside the 95% of cases you mention.
  • bantel_catbantel_cat Posts: 969 ✭✭✭✭✭
    Awesome work! Thanks.
    Any interesting outliers in the cars you tested, good and bad? 
  • RexKwonDoRexKwonDo Posts: 63 ✭✭✭
    Nacho101 said:
    Nice thanks, interesting to see how much of an impact 0-60 time has over grip. Im guessing the time must still vary considerably when comparing cars with same 0-60 but huge difference in grip/weight - that would fall outside the 95% of cases you mention.
    Yes, if cars have the same 0-60 but a big difference in grip it can definitely create variation in the outcome.  The full equations take the grip into account, but the graph here is just the 0-60 impact by itself. 

    Looking again, using the full equation I guess 97% of my cars met the 0.5 second threshold and 72% were within 0.2 seconds.  No sure how I managed to fat-finger that one so well. I've gone back and edited the original post to correct the number.

    For medium clearance cars the grip actually adds more information to the equation than 0-60 does.  For low and high clearance it adds a lot less.  
  • RexKwonDoRexKwonDo Posts: 63 ✭✭✭
    Awesome work! Thanks.
    Any interesting outliers in the cars you tested, good and bad? 
    For this course, the Honda Beat, BMW Z1, and Caterham 21 were all more than 0.5 seconds faster than predicted.  The Lancia Fulvia, Mazda RX-7 (1978), Camaro Convertible, and Caterham CSR were all more than 0.5 seconds slower than predicted.  All other cars that I tested fit inside of the +/- 0.5 seconds.  

    Surprises for me included the Camaro Convertible.  It's still fast, but based on the equation it should have been even faster.  And then the AMG S 55 was my second fastest B car, only 0.03 seconds behind the Camaro.  With a grip of 82 I didn't think it had any business being in there, but that car is always surprisingly good.  
  • OzzmanOzzman Posts: 1,714 ✭✭✭✭✭

    Very interesting one again. I guess city streets medium is where grip matters much more, the small one is more about speed.

    Surprised to see that med and high ground clearance can make a difference. Even more surprising is that the chart above (and the equations also) suggests that med ground clearance is faster (although not by much) than high.

    Can't imagine how much time is needed to do so much testing. :)

  • HeissRodHeissRod Posts: 6,904 ✭✭✭✭✭
    Medium ground clearance appears faster because vehicles with high ground clearance tend to be slower.  That's my theory anyways.
  • UltimateUltimate Posts: 777 ✭✭✭✭✭
    HeissRod said:
    Medium ground clearance appears faster because vehicles with high ground clearance tend to be slower.  That's my theory anyways.
    I always thought that cars with high height in real life had a real possibility of rolling if they took corners too quickly, might that translate in the game as having worse handling or the AI slowing high height cars down more for cornering. One for the myth-busters 🤷‍♂️
  • OzzmanOzzman Posts: 1,714 ✭✭✭✭✭
    Ultimate said:
    HeissRod said:
    Medium ground clearance appears faster because vehicles with high ground clearance tend to be slower.  That's my theory anyways.
    I always thought that cars with high height in real life had a real possibility of rolling if they took corners too quickly, might that translate in the game as having worse handling or the AI slowing high height cars down more for cornering. One for the myth-busters 🤷‍♂️
    I don't think the model is that complicated. I think (as HeissRod said) it's just reflecting that higher cars tend to have worse stats. To be more precise, their hidden stats (like MRA) tend to be worse. But as always, only Hutch can know for sure.
  • UltimateUltimate Posts: 777 ✭✭✭✭✭
    Sorry I went a bit off topic there, I was just thinking how real life physics would translate into TD when cars obviously cannot roll as they might in real life doing sharp corners at speed.
  • RexKwonDoRexKwonDo Posts: 63 ✭✭✭
    Ozzman said:

    Can't imagine how much time is needed to do so much testing. :)

    @Ozzman Thanks.  I usually pick a campaign course that has multiple tracks of interest for me. That way I collect data for lots of future analysis at the same time as something I'm interested in right now.  Then during boring conference calls I fire up the game and record times.  But the bigger my garage gets the more time it takes so I largely ignore cars from B-F that aren't maxed out. 
  • David_FookDavid_Fook Posts: 2,713 ✭✭✭✭✭
    @RexKwonDo have you thought about crowdsourcing the times? If you created a spreadsheet we could all access im sure we’d all fill in some times of cars we were interested in. This way you’d get a bigger range, and more high-end cars 
  • MoogMoog Posts: 459 ✭✭✭✭
    RexKwonDo said:
    Ozzman said:

    Can't imagine how much time is needed to do so much testing. :)

    @Ozzman Thanks.  I usually pick a campaign course that has multiple tracks of interest for me. That way I collect data for lots of future analysis at the same time as something I'm interested in right now.  Then during boring conference calls I fire up the game and record times.  But the bigger my garage gets the more time it takes so I largely ignore cars from B-F that aren't maxed out. 
    Thanks for the info and analysis, an interesting read!
    Regarding the collection of times, I'm sure a few of us wouldn't mind helping you crowd source times to help with your research. A number of us keep track of times, although historical data gets wiped when physics updates change, so please do shout.

  • Hutch_TimHutch_Tim Posts: 666 admin
    Ultimate said:
    HeissRod said:
    Medium ground clearance appears faster because vehicles with high ground clearance tend to be slower.  That's my theory anyways.
    I always thought that cars with high height in real life had a real possibility of rolling if they took corners too quickly, might that translate in the game as having worse handling or the AI slowing high height cars down more for cornering. One for the myth-busters 🤷‍♂️
    This is exactly the kind of thing that should get translated into worse handling.

    In terms of City Streets Short vs. Medium, back when they had an indistinguishable name you could tell it was Short if the highlighted blue tip was only on 0-60; for Medium it highlights both 0-60 and handling!
  • MoogMoog Posts: 459 ✭✭✭✭
    Hutch_Tim said:
    This is exactly the kind of thing that should get translated into worse handling.
    That's true of all height cars though, right? If you turn them too fast they will flip as lateral forces and inertia exceed the grip; it's just easier to demonstrate on cars with a higher centre of gravity. If you're factoring that into the handling score then it should also be coupled with the wheel base.
  • MoogMoog Posts: 459 ✭✭✭✭
    HeissRod said:
    We've discussed this in the past.  Crowd-sourcing a database of times for cars would ruin the game.  All a user would have to do is reference the database to see exactly which cars they should use and where exactly they will win.  Then everyone does the same and we have no creativity or individual testing.
    I agree with you, I'm talking about crowd sourcing to aid his research, it doesn't need to be made publicly available.
  • RexKwonDoRexKwonDo Posts: 63 ✭✭✭
    Moog said:
    I agree with you, I'm talking about crowd sourcing to aid his research, it doesn't need to be made publicly available.
    I appreciate the willingness!  I generally have enough data points to get a good model of each track and not be worried about the error that could come with low number of data points.  On occasion there is a group I am short on, such as Slick tired cars.  (My garage is also notably short on maxed out legendaries and epics, so it is hard for me to discern if they behave differently as a group.  So far, for every test I've run, my legendaries just show up as unusual data points, but still very aligned with the models I generate.)

    Perhaps I could ask the group for specific data at times when I'm researching a particular topic in which I see a gap in my own garage?  For example, I'm halfway through some data collection on drag times on wet asphalt to compare with dry asphalt.  But, I only have one Slick car so maybe the group would be willing to share specific track times for their Slick cars.  In those cases I could supply a link to a google sheet for those that were willing to share that specific set of data.  
  • HeissRodHeissRod Posts: 6,904 ✭✭✭✭✭
    It would need to be a submission form of some sort, so the sheet could not be used as a point of reference, except by you.
  • RexKwonDoRexKwonDo Posts: 63 ✭✭✭
    HeissRod said:
    It would need to be a submission form of some sort, so the sheet could not be used as a point of reference, except by you.
    @HeissRod Would something like this work?  I just threw it together quickly, but just wanted to test the idea.

    https://docs.google.com/forms/d/e/1FAIpQLSfD5tD2sdY_gJWkNh_1SbJaDZhdkcF6Y0rzRJ3CVn9Qfgcw0A/viewform?vc=0&c=0&w=1&usp=mail_form_link
  • bantel_catbantel_cat Posts: 969 ✭✭✭✭✭
    RexKwonDo said:
    HeissRod said:
    It would need to be a submission form of some sort, so the sheet could not be used as a point of reference, except by you.
    @HeissRod Would something like this work?  I just threw it together quickly, but just wanted to test the idea.

    https://docs.google.com/forms/d/e/1FAIpQLSfD5tD2sdY_gJWkNh_1SbJaDZhdkcF6Y0rzRJ3CVn9Qfgcw0A/viewform?vc=0&c=0&w=1&usp=mail_form_link
    Yep that works well.
    Submitted times for the 911 RSR and the XJS as those are the only Slick cars I have.

    Interesting that the Jag is quicker on the drags so I assume the lower power/weight ratio helps get more traction on the ground.
    Be interesting to see what the regression analysis will say, hopefully, the sample size will be big enough.

  • hillclimberhillclimber Posts: 2,233 ✭✭✭✭✭
    am I missing something or should the tuning stage be added?
  • lonelyterroristlonelyterrorist Posts: 133 ✭✭✭
    RexKwonDo said:
    Awesome work! Thanks.
    Any interesting outliers in the cars you tested, good and bad? 
    For this course, the Honda Beat, BMW Z1, and Caterham 21 were all more than 0.5 seconds faster than predicted.
    Would this have anything to do with the 0-30/40/50 times? Beat is great off the line to 40 but peters out at much past that. 
  • RexKwonDoRexKwonDo Posts: 63 ✭✭✭
    am I missing something or should the tuning stage be added?
    I'm not sure if I understand the question right, but let me take a stab at it.  If we are talking about the Wet vs Dry all I need are course times because I will end up generalizing a percentage impact for a category such as 4WD-PER cars.  In the past I've found that the percentage impact stays pretty consistent as I upgrade the cars.  If we are talking about any other analysis then yes, the tune matters, but not as a piece of data on it's own. Rather because it dictates the other stats. So it's more important for me to know peak power, torque, weight, 0-60, top speed, etc. 
  • RexKwonDoRexKwonDo Posts: 63 ✭✭✭
    lonelyterrorist said:

    Would this have anything to do with the 0-30/40/50 times? Beat is great off the line to 40 but peters out at much past that. 
    @lonelyterrorist Quite possible. The Beat has sometimes been difficult for me to pin down with stats as it tends to break the rules all the other cars are held to. Might very well be related to the 0-30 time.
  • LittleEnosBurdetteLittleEnosBurdette Posts: 3,076 ✭✭✭✭✭
    All very interesting stuff, too technical for me though. Just to add though that because a car shares ride height it doesn’t necessarily perform on city streets the same as other cars with the same height, there must be other factors. 
  • 43MK443MK4 Posts: 1,910 ✭✭✭✭✭
    @RexKwonDo Very interesting work! some time ago you made a post on IG forum regarding impact of rain on twisty road for different tires.
    I was wondering if you would be interested and could have a look on race times on some wet track in relation to 3 groups of cars:
    1. No ABS and no TC
    2. ABS but no TC
    3. ABS and TC

    Wet twisty road could be used since you already have plenty on data for that.
    im really curious how big/small impact ABS and TC really have.

  • HeissRodHeissRod Posts: 6,904 ✭✭✭✭✭
    edited February 2019
    Keep in mind that 4WD cars with performance tires just received a nerf on wet asphalt, so his data could have been affected 
  • willcf15willcf15 Posts: 359 ✭✭✭✭
    edited February 2019
    Hey buddy, this is pretty neat, thanks for posting!

    Any chance you would consider using RQ as the independent variable in this or a future study? I do understand why 0-60 was used, because it most directly compensates for slowing down and speeding back up over the bumps, so it makes sense that it would be the best predictor. It's nice to be able to guess when you'll be able to compensate for ride height with brute force.  However, I think I'd be able to draw some interesting conclusions from RQ vs height and CSS time, like essentially how much city streets ability "costs" from an RQ perspective.
  • RexKwonDoRexKwonDo Posts: 63 ✭✭✭
    @43MK4 Sounds interesting.  I'll put some thought into it.  Once my work settles down a little I will get back on some of the analysis. 
  • RexKwonDoRexKwonDo Posts: 63 ✭✭✭
    @willcf15 Thanks for the thoughts.  It's intriguing to think about whether we could assign a "cost" to a lower RQ here.  

    I've avoided RQ as a predictor of course times for a couple of reasons. First, it appears to be sort of a subjective number (as evidenced by the RQ changes that happen at updates due to some cars being misclassified) so I don't trust it to be reliable. Second, just from a data perspective, RQ almost always flags in my model for high multicollinearity.  In other words, RQ highly correlates with 0-60, grip, and Peak Power (and even height if it's in combination with the others) - together they all predict RQ rather than RQ being the factor that does the predicting of results. 

    However, it would be nice if you could just have a simple RQ gauge to go by. It would really simplify things.  Looking at RQ by itself, there is a definable relationship between RQ vs CSS results.  Here are three charts --one for each ground clearance-- if you want to generalize a "cost" to the lower RQ on the course. 
Sign In or Register to comment.