The Hills

[EnigmaMTC]

is getting past 750 on generation 3 [from random seed] normal?

Code: Select all

eNqzP8lm8iotLc3+2Cu3pgYGBvvdDIKmIPrbZzlRZ2Nj+wMMHUbGQPqp2IyOA0DxS0YC72fOnGl/zkTEejNQ/HCpYCDPmTMObA+WHfUAyh+PnWvXc+aM/T/Ozvduaen27wSMLlvOnGX/f1Xqxs9A9XdSNeV0gPI/JDidLxsb/wcCB47Gm8KtB1fbv7xT2GgMFRP+eJRDm2+b/WP2jKRiY2MGBgZW+0cd7sdcXL3s39189ALoBqAYm/0nu4PKHEfu239TO757M0Sv/etlD27PvzXR/m2Bln4kWB2jA4/86s+2357bv1Bz3Q4UA9shmHNsup5dvf3emVlBZ86chdhbcmbjz40d9p/7c6fOnDmLwfG3HgZexlEAx8566VjVwDDQ3QD9B511

[TechSY730]

is getting past 750 on generation 3 [from random seed] normal?

No, but it most certainly is not impossible. With all these people trying to improve on The Hills, chances are, at least one person would see a pretty good car so early from random seeds. You were the lucky guy.

[EnigmaMTC]

yeah... 3 hours of simulation later, and the score from this car has not been improved... the closest a mutation of it got to the original score is about 678.

[meh]

So, I've been putzing with a number of these codes, but there seems to be differences between the flash implementation of whatever everyone else is using (probably windows) and my own flash implentations (Ubuntu and OSX). Cars that are getting great scores for you guys are dying in the 500s for me, and cars getting 1200 in Linux are dying in the 600s for OSX...

My highest score is 1498.7 exploiting a bug in the bike building code (and a couple weeks of letting it grind):

Code: Select all

eNqzf7Z/wqzLfXPtd84EgVn2b7TqJkrbOdifXr7GSP3kEfu3psxrlqWl2z/LLRY5/P2C/ZNa+5JHWpz2d1meJW0vr7B/5rGu8wIDg/3zv4rf8t69t390ymj9JsY4+1ebtS4vitjvEGjSMHfuSRH7t1G7MmovfbJ/fUWGTc3jgf371RzCEZNL/wOBg3hyMbsFQ7X9ky4/gy3K9QxA4MA4/WT7yaAj9nePT+Dvdp4KVse2NFrdXZnH/umiELcGBgaQmH1hUMQCvx1M9m/nxrMnKSkAtTI6CHoINz92UrP/fCigv6OjAyjG6sB49uj7Ttcp9j8YYIDNQeiIg+LF//NhYmA7RDJ3PXzPy2P/tUrmckRSI0PiqsUMTyO+MXBPOs3A7HSKYZrpJwaZhH6Gat8UhuD7Kgy8nrEMitEWDAnrhRmij6cy1HKwMyjaH2foOn6UIcRMGmgsMwCeCZWH

I've gotten 1200, but only on Linux:

Code: Select all

eNqz/8AABvbvHvJEC9vJ219M1W5fJG5q/6LbeWojm6o9VN6BReHmfR2gOpj6p4f6YzS+HrJ/kZNkkdez0P6zj6SdqPAZ++f3//kqzpxl/ykqdinTqlX2J38Jl9+cOdP+Yqe597HlH+wfHZsjL7Vqtf3HDY8yN3zwARrF5CCp+DvERVfC/sUWywiBJF+gGJsDq8/5GUc9De2/5U3v9b9v8x8I7P+X+h7Y2VZk/+m39bkUmQlAdcwOwsseHuww2mT/8Kws59+798Dqfpsasgrr7bZ//zZ/it67d0B1jA7CylmNCW537L/3e6XHu9wEirHa/16k80DA7ZT9DwYYYHLgMouombg+0/65xSklP6DeNyFZDPLXDzFsn8TD8KhqAsOrX/UMTusZGX6Z3mKo59NgYC9vYeAMs2SwuLGWwfxxBYPilLUMbfs9GOJOvmFIaACZyQwARwWOTw==


I keep getting a wild hair to figure out which flash implementation is more accurate. I figure that I am looking at a floating point rounding issue that is causing this.

Either way, the weight and wheelbase limitations made by earlier sections of terrain are exploited by later sections of the track.

[Lotharsson]

So, I've been putzing with a number of these codes, but there seems to be differences between the flash implementation of whatever everyone else is using (probably windows) and my own flash implentations (Ubuntu and OSX). Cars that are getting great scores for you guys are dying in the 500s for me, and cars getting 1200 in Linux are dying in the 600s for OSX...

I found that some cars performed differently on the same machine depending on whether the Flash object was hosted directly in an ActiveX container or by a browser object, so it's not just OS platform that affects it. And without changing anything else, changing the least significant bit of a single chromosome value can be enough to dramatically change scores on a track like The Hills.

My highest score is 1498.7 exploiting a bug in the bike building code (and a couple weeks of letting it grind):

That's a fantastic score, even if it doesn't work on all platforms!

Just tried it on Windows 7 using Chrome and it gets trapped at 1075.7 These platform differences are a PITA. Do you get the same score on both Linux and OSX?

But I don't know why I didn't think of using my Sisyphus negative mass bug car on the Hills before - great idea!

I've gotten 1200, but only on Linux:

Yep, only 788.6 on Win7/Chrome

[meh]

I found that some cars performed differently on the same machine depending on whether the Flash object was hosted directly in an ActiveX container or by a browser object, so it's not just OS platform that affects it. And without changing anything else, changing the least significant bit of a single chromosome value can be enough to dramatically change scores on a track like The Hills.

Oh yeah, I'd left them running on test boxes and would glance back to see how they would be doing while running a compile. The 1200 car had to have a certain wheelbase/wheel based on some of the earlier cars you guys had. I took that length and tried to build mass in front of the center of the front wheel. At 1200 it flops and hits right at the suspension point killing the vehicle.

That's a fantastic score, even if it doesn't work on all platforms!

Yeah, but with the weight bug, I knew I was cheating. It was a matter of getting the right shape to avoid getting hit, and then letting it match a wheel size to the mass/terrain. The end of the hills is at that point and that's what I wanted to see. I wanted to be sure there wasn't a sheer face or worse an overhang that would make it impossible to complete.

Just tried it on Windows 7 using Chrome and it gets trapped at 1075.7 These platform differences are a PITA. Do you get the same score on both Linux and OSX?

Didn't do the single wheel on OSX yet. It was working on a home server, but I'll bring it in tomorrow (WFH today).

But I don't know why I didn't think of using my Sisyphus negative mass bug car on the Hills before - great idea!

Thanks, admittedly, it is kinda cheating and not worthy of being a leader for the hills, but it did allow me to map out the hills in my head and think about a basis for a design.

I've gotten 1200, but only on Linux:

Yep, only 788.6 on Win7/Chrome

Better than that thing gets on OSX. I'm going to look at your Sisyphus code and see what that does on sisyphus and other tracks on the various platforms it might give me some clues. It might give me an idea of what the rounding errors do to performance. I still believe a "fish" shape will get the furthest. The 1200 car is very entertaining on Linux with multiple flops and a tendency to have to drag itself over obstacles after the 750 mark. The 788 mark makes me want to give it more mass up front and maybe a larger font tire (I tweak around some issues just for grins).

[meh]

Do you get the same score on both Linux and OSX?

That code got 1019 on OSX this morning. I'm letting it run for the rest of the day.

[honnza]

W7/FF4. No glitch.

[meh]

That code got 1019 on OSX this morning. I'm letting it run for the rest of the day.

Code: Select all

eNqzf1WUYCrvFWm/cyYIzLJ/o1U3UdrOwf74xAfTp93ZZv/WlHnNsrR0+2e5xSKHv1+wf5i2dILHhHz7uyzPkraXV9g/81jXeYGBwf5F/B+L0gd+9o9OGa3fxBhn/2qz1uVFEfsdnE04PtrfOWH/NmpXRu2lT/brj2euyePNsX+/mkM4YnLpfyBwEE8uZrdgqLZ/0uVnsEW5ngEIHBinn2w/GXTE/u7xCfzdzlPB6tiWRqu7K/PYP10U4tbAwAASsy8Miljgt4PJ/tN626isVauAWhkdBHJvzVJ2U7L/fCigv6OjAyjG6sB49uj7Ttcp9j8YYIDNQeiIg+LF//NhYmA7RDJ3PXzPy2P/tUrmckRSI4NQhy/D7W0/GZJVZRkmPNjK8Eajn0EmoZ+B61UhQ/B9FYZZO00ZZt8uYkhYL8ygt9+H4R5bNIOi/XGG02JNDCFm0kBjmQEG3pet

[Lotharsson]

Bit of a breakthrough

The Hills 1207.7.PNG (9.34 KiB) Viewed 1644 times

Then it gets stuck again...