Drivers2 Ratings Test

[JoyfulPanda]

Hey ya'll - I was working on ratings for a carset I'm using and wanted to get to the bottom of the effect of the Drivers2 ratings.  I thought some people might find it useful.

THE EXPERIMENT
36 races (9 each) on 4 speedway tracks:
* Fontana
* Indianapolis 2000
* Michigan 2000
* Pocono

(Thanks to Pavel for the great work on these tracks).

Accelerated Time races of 25% length (long enough to see the effects, but short enough that DNF's didn't pollute the results too much).  Damage set to Arcade to reduce crash DNFs, since I was not testing Aggression in this experiment.

ICR2 is version 1.0.2 on DosBox on a Mac.

30 cars.

Each car was initially set to:
500 501 500 501 500 501 1 2

And then modified by changing a single value, as indicated by the name.  For example, listed below, "Trac510" means that car's ratings were:
500 501 510 511 500 501 1 2

With those parameters, here are the results:

CAR      STR   AVF   SD
Trac590   9      1.12   0.33
Trac570   10    1.94   0.76
Trac550   12    3.03   0.72
Trac530   14    3.67   0.53
Trac510   15    7.58   3.65
---
Powr950   1    8.42    4.66
Powr850   3    9.26    5.13
Powr650   7   10.44   4.71
Powr750   5   10.56   5.19
Drag050   2   11.29   5.67
Drag150   4   12.89   5.63
Powr550  11   13.03   5.70
Powr350   23 14.31   4.88
Powr450  18  14.38   5.54
Drag350   8   15.39   4.70
Drag250   6   15.42   4.78
Drag450  13  15.52   4.66
Drag550  20  16.35   5.68
Powr050  29  16.86   4.69    
Powr250  25  16.91   4.69
Powr150  27  17.63   3.48
Drag750  26  18.33   4.84
Drag650  24  18.34   4.86
Drag950  30  19.37   4.30
Drag850  28  19.62   5.49
---
Trac490  16   20.26   2.92
Trac470  17   24.69   3.07
Trac450  19   26.15   1.23
Trac430  21   27.13   1.38
Trac410  22   27.97   1.40


NOTES/OBSERVATIONS
Qualifying is determined EXACTLY by the average of Power, Traction, and (1000-Drag), so starting grid was always exactly the same - and for example, Power 550 had the exact same qualifying speed as Traction 550 and Drag 450 - with the tie appearing to be broken by driver order.

I separated out the Traction cars to emphasize the observation that traction rating overpowers all else.  I initially started this experiment with Traction spread just like Power/Drag, but it was so clear that traction had such a large impact that I condensed it to a smaller range of variance.  However, even a ten point boost in traction outperformed a 450 point boost in power, while starting from much farther back in the grid.  And the reverse was also true, with a ten point loss in traction sending a car all the way to the back.

Increases in power outperformed reduction in drag, but at the same time, reduction in power outperformed increases in drag.  While one car being out of place in the rankings can be attributed to noise, a group of five cars outperforming another group of five cars that was expected to be similar can't.  We would need further testing, but either:
* Drag and/or power is somehow non-linear - for example, could be plausible that lowering drag beyond a certain point has limited impact, while raising it to higher values has a much greater impact.
* The starting grid position has some lasting impact (seems less plausible)
* There are things hardcoded into the driver slot that are not captured by these editable driver rating values
* Chassis/engine/tire has an impact.  I did not pay attention to these values at all when conducting the experiment, as I assumed they were only cosmetic.


CONCLUSIONS & FURTHER QUESTIONS
When setting up your carsets, use traction as the car's race pace, and use power/drag as adjustments for qualifying pace relative to their race pace.  Can create strong qualifiers that fade during races or vice-versa.

I'm not confident in our ability to tweak these values to make drivers stronger weaker on ovals vs street tracks - more testing is needed (next up for me would be street tracks), but based on how little impact drag has even on large ovals, it seems unlikely.

As mentioned above, might need some additional testing to tease out the impact of engine/chassis/tires, if there is one.


Hope people find this valuable!

[JR82]

I have been doing my own head in trying to figure this out for a number of yrs & have found similar to what u mention here. Also to see if/any of these values have an impact on overtaking for example. I guess a way to properly test it would be to find each of the 4 values as a base that starts off the same performance as each other (which i have yet to figure out let alone properly test). by that i mean with a power rating of say 500, which traction value will match that for race pace? from here we may be able to figure out if we can have "oval masters" like Penske most seasons ('97 for example & Brack '01) for example & "road course masters" that are elite on road courses but rubbish on ovals & vise versa etc. Being able to impact reliability would also be great but not sure if any of the above is possible.

[JoyfulPanda]

In some limited, less formal testing, I have observed what some other people have - that lower aggression values will make more overtakes than those with high values.  Which is why I set aggression to 1 / 2 for this experiment.

And I'd love to get to a point of like X Power = Y Traction on speedways or something, but since Traction is so much more impactful it seems almost impossible to get to the level of precision we'd need here...

At the moment I'm resigned to making multiple carsets (same cars, same drivers, just different Drivers2 files) and swapping them out for different course types to get those Oval/Road differences you mentioned

[Tony_D 26nz]

Wow, what great analysis & report. Thanks Joyful!

I enjoy doing this sort of stuff myself, although my skills are order of magnitude behind yours...

[samsepi0l]

As mentioned above, might need some additional testing to tease out the impact of engine/chassis/tires, if there is one.

Interested in understanding this. When I do my carsets I give all cars the same chassis (reynard) and engine (honda) but different tires for graphical reasons.

I name the car type (lola, reynard, march, eagle, whatever) and engine name in the text of the opponents file, so it comes up correct when you view the opponents in the spinning screen.

[CowtownBob]

On speedways, Traction will be the major determinant though Power will have a relatively increased effect for the first lap or two while cars come up to speed (Power = acceleration). But that's not going to hold for other track types.

None of the values are cosmetic only; there would be no reason to have them in the file (no good reason anyway, I can think of a bad reason or two). Back in the day people thought they did not effect the player car, which is also wrong.

[checkpoint10]

None of the values are cosmetic only; there would be no reason to have them in the file (no good reason anyway, I can think of a bad reason or two). Back in the day people thought they did not effect the player car, which is also wrong.

Are you saying that if you change the power, traction, drag, etc. in drivers2.txt for the player's car, it would actually affect the player's car?

[JoyfulPanda]

In other less formal tests I have seen traction completely dominate over all other values. The closest to another value mattering was on street courses - where power values of 900-999 could almost match a traction car with 600-699 values. On road courses and non-speedway ovals, saw similar splits to what I saw on the speedways. But need to do a more formal run to verify

[checkpoint10]

In other less formal tests I have seen traction completely dominate over all other values. The closest to another value mattering was on street courses - where power values of 900-999 could almost match a traction car with 600-699 values. On road courses and non-speedway ovals, saw similar splits to what I saw on the speedways. But need to do a more formal run to verify

Can you analyze replays of AI cars, comparing the times going through different sectors, to see where power/traction/drag have the most influence? I think there are replay analyzers that can provide such data. Maybe you would also need to run only one AI car to ensure it is following the normal racing line and not the passing lines.

[CowtownBob]

Are you saying that if you change the power, traction, drag, etc. in drivers2.txt for the player's car, it would actually affect the player's car?

Yup. Power is the easy one to test; set all cars to a high value, say 900 - 901, start a race without qualifying or pace lap so you're on the back row; you will easily keep up with the field. Then change the player car to 100 - 101; you'll be left in the dust. Drag could probably use a similar method. Testing the other values might not be as easy because they're dealing with cornering speeds, overtaking, etc. so you might have to go in and analyze replay files for values to compare.