3hr Heatmap Marathon Map Test

[Ravidge]

The big blurry red blobs are great for an overview and pretty pictures, they look great, super nice.
But for actual information gathering... to me as a mapper I have to look really hard to find what's relevant.
The pictures get prettier the blurrier the blobs get, and the larger the red and yellows get, but the ability to read anything from them just hits rock bottom very quickly.

I'll admit I've never really seen these kinds of heatmaps as particularly useful for mappers. The shape of the blobs always seem so arbitrarily defined, not being shaped to the rooms, spilling over into a completely irrelevant tunnel that happened to be next to a CP, but in reality it sees very little action, for example.
This is most likely due to a small sample size, which makes these huuuge green clouds that span multiple rooms but don't really tell you anything useful.

For this to be useful for testing purposes and not just finished maps that are in rotation on popular servers it needs to cater to low data/small sample size readability. A map in progress can see as little as 1-3 tests before the next version, that's at best 1½ hours of data. Some of it probably collected with less than 24 players.

Will a heatmap help me there? Probably not, because everything will be blue or have these massive diffuse green gas clouds covering everything.

I took the first image you posted in this thread and had a go in photoshop with a few channel mixers and saturation filters and came up with this to help me find what was actually worth looking at in that heatmap

1) I removed the blue, it might as well not be there if it's not going to tell me anything.
2) I made the edges AND colors much sharper/intense.
3) I removed green gas, if the green wasn't intense enough it got culled completely, it's just irrelevant random noise deaths. The level of intensity that you deem relevant is probably the hardest thing to tweak.
I did not draw anything, I only tweaked settings and adjustment layers. It also looks like ass.

I found some green spots that were worth looking at, but being unable to tell if they're red or blue team deaths I would have to look at the demo for anything further.
The spot between mid-2nd is probably the most interesting, as it seems to be in a different location depending on if your team is stomping or not. Because clearly one team was winning most of the rounds. Or it's because of class composition of the teams?

I'm not going to read further into what the heatmap might mean, but this definitely helps me isolate a few key areas at a quick glance. And let's not forget that maybe I was expecting green in some areas where there's none? Now that I have these spots I can focus my attention to them when watching the demo.
As a mapper I don't really need the full gradient of green-yellow-red. Just give me those 3 colors, the in-between shades don't really matter to me. (maybe add blue if it's going to be useful).

I've never played glassworks, so I don't know anything about its gameplay or layout


Maybe this whole post is a load of bullshit. But maybe you can see what I'd like in a heatmap tool if it's intended to aid map development. I want well defined easy to read hotspots of different colors. Remove noise from the data and show me what's relevant.
That makes for really ugly pictures though.

 

[InstantMuffin]

Smoothing set to 12/100 and the min score is at 20%.

If you want the base temperature can be adjustable, so that it starts at green for example.

 

[littleedge]

Yeah, I've never really understood why these bubbles-of-color heatmaps are useful. The CS:GO ones show me where people were when they died - the exact spot. These ones show me the general area of where people died (and includes areas where people didn't actually die due to smoothing or whatever it is), a much less useful thing.

 

[InstantMuffin]

Yeah, I've never really understood why these bubbles-of-color heatmaps are useful. The CS:GO ones show me where people were when they died - the exact spot. These ones show me the general area of where people died (and includes areas where people didn't actually die due to smoothing or whatever it is), a much less useful thing.

I think it is rather obvious that smoothing is adjustable and unfortunately the only way of compensating lack of data. Heatmaps a tool, the tool to create them will be a tool. There will be settings to adjust all kinds of stuff. Just going ahead and throwing an image at someone is zero help. That's why I tried to upload as many heatmaps as possible with different settings to give an overview.

 

[Ravidge]

Smoothing set to 12/100 and the min score is at 20%.

If you want the base temperature can be adjustable, so that it starts at green for example.

That already looks a lot more useful, at least in my eyes. But I'm just one guy
With that I can definitely separate out signal from noise. A lot of the time, knowing where to look can make a world of difference when watching gameplay on your map.

Thumbs up.

 

[WolfKit]

What does it look like if you display only red deaths, or only blu deaths?

 

[YM]

Still don't understand this minimum % business, or really why there's much smoothing going on. I'm rapidly being won over by the CSGO style of clear and resolvable points.

 

[InstantMuffin]

Still don't understand this minimum % business, or really why there's much smoothing going on. I'm rapidly being won over by the CSGO style of clear and resolvable points.

Because that needs a much larger dataset to be effective.
Alternatively, just turn smoothing down. Points will just be round and not squared. I don't get the ranting from your side. Obviously you won't be using this either way. Just be happy for the rest of the people.

What does it look like if you display only red deaths, or only blu deaths?

I actually forgot to do that.
There's a few other things I have noticed now that I have a bigger dataset.
I'm recording the weaponid to determine the kills. While this enum has an index for sentrykills (defined by Valve), Valve doesn't use it. So minor testing needs to be done to properly route around this. I'd prefer to do this with Frozen when he is back.
I also like to record each player teleport then.

 

[Ravidge]

What about an option to overlay the specific death position 'pixels' on top of the color heatmap, if you want them? As in using both the blob and pixel method of heatmap at the same time.

Showing exactly the number of dots in each area. Following on the last picture you gave me, could there be something like this, where it shows me exactly the data it's using to generate the blobs. (these are 2x2 white dots with a 1x1 blue border)

I'm sure there should be more dots, in which case 1px dots would be better. It sort of breaks down if you get too many dots though. But for low sample map tests I think it maybe could be cool? It lets you get a grasp of what created the blobs and their size and shape becomes less of a mystery.

It's hard to say. I just don't know the number of deaths you used to generate those blobs in that image (which also would be great to know, like "xxxx out of yyyy kills shown as heatmap data."). I know you said something about 20% but I don't really know what that means.

I wouldn't want this for every heatmap generated, but I think it could be interesting to have as an option?
I would like to hear what other people think though, maybe I'm spouting nonsense again.
Or perhaps it's just a bad idea altogether, it would just produce garbage images?

 

[YM]

Because that needs a much larger dataset to be effective.

If you plotted each death on a 32 unit grid, what would that look like?

You could then dynamically scale the grid used depending on the amount of data. When a map has only been played a tiny amount -> 64 unit grid, then 32, then 16 for a map that's got thousands of datapoints.

I'm ranting because I'd rather have specific data, than a pretty heatmap.

 

[InstantMuffin]

If you plotted each death on a 32 unit grid, what would that look like?[...]

Pixelated.

[...]I'm ranting because I'd rather have specific data, than a pretty heatmap.

You wouldn't. Basically you'd have squares instead of circles, as said before. The approach as it is is actually more sophisticated.
Don't like smoothing, turn it down. There's no reason to prefer square-sized blobs over round ones.
Unfortunately there's nothing more to say about it without repeating myself.

What about an option to overlay the specific death position 'pixels' on top of the color heatmap, if you want them? As in using both the blob and pixel method of heatmap at the same time.

Showing exactly the number of dots in each area. Following on the last picture you gave me, could there be something like this, where it shows me exactly the data it's using to generate the blobs. (these are 2x2 white dots with a 1x1 blue border)

I'm sure there should be more dots, in which case 1px dots would be better. It sort of breaks down if you get too many dots though. But for low sample map tests I think it maybe could be cool? It lets you get a grasp of what created the blobs and their size and shape becomes less of a mystery.

It's hard to say. I just don't know the number of deaths you used to generate those blobs in that image (which also would be great to know, like "xxxx out of yyyy kills shown as heatmap data."). I know you said something about 20% but I don't really know what that means.

I wouldn't want this for every heatmap generated, but I think it could be interesting to have as an option?
I would like to hear what other people think though, maybe I'm spouting nonsense again.
Or perhaps it's just a bad idea altogether, it would just produce garbage images?

The overlay thing is an interesting idea.
But I think people are forgetting that the simple "one pixel for each kill" thing has been included ever since, simply for the purpose of convenient calibration.

20% was the min score of the pixel in comparison to the max score of a pixel (the "hottest" pixel") that it should have to be colored.

 

[YM]

Before I finally give up and shut up, will you just post a picture of no smoothing at all, and none of this 20% score business. I want to see every kill, no smoothing.

 

[Fruity Snacks]

Post the original dots version, muffin.

 

[InstantMuffin]

I'll do that tomorrow. Can't do anything at the moment, I am unpacking and hashing content files for a different project I'm working on (which I have finished today ).
Need an SSD. -.-

 

[Crash]

Missed this edit back on page 2 the first time!

Min score is set to be 5% of the max score now. (Will be configurable)

These ones are definitely getting there! I think it's just a matter of fine tuning things.

 

[yoghurt]

by using the smoothing you make it look like you have a lot of data when you don't. It really just blurs the result in my opinion.

 

[henke37]

Can we just get the dataset already?

 

[Crash]

Can we just get the dataset already?

Did you miss the part where Muffin said he would get to it when he could? Don't be an ass to someone who is working hard to create something of great value to our community.

 

[Fish 2.0]

The main issue here that people aren't really conveying to well is that the smoothing is giving us false data.

Sure, it looks like a heatmap but it's giving us information that isn't true. I could point at pretty much any coloured pixel on that heatmap and there wouldn't have been a kill in that exact location. It generalises to such an extent that it is giving information that isn't telling us anything. I point at those images and say "there will be a lot of traffic around this area and deaths here" because it's common sense.

What we're after is something that can give us a detailed image of what we already know, not repeat what we already know. I guess that's why people are asking for a extremely specific version.

 

[fauks]

A map like the attached might be useful for map designers with fewer data to go on. Filter by team. Optional filter by class. The dot represents killer location, the line terminus is the killed player.
Magenta/cyan colours for clarity, rather than red/blue.
The example shown below is a quickie.
I just think this might help identify overpowered locations.