3hr Heatmap Marathon Map Test

Crash

func_nerd
aa
Mar 1, 2010
3,315
5,499
3 Hour Heatmap Plugin Marathon Test

So as Frozen is busy on super important, super secret train related business, and as my map Glassworks was chosen in the poll, I'm going to be taking over this event. The gist of things is that we need a lot of data on one map to test the plugin/ calibrate things. That's where you nerds come in!

If you're unfamiliar with the heatmap plugin, Frozen did a good write up here, so go read up on it!

Here's the skinny:

Who:
All you nerds!

What:
Glassworks

When:
Thursday June 19th @ 5PM EST Until 8PM EST

Where:
Due to issues with the plugin and the US server, we will be running the test on the EU server
TF2Maps.net Official EU Server: 91.121.155.109:27015

Why:
Because we need a lot of data. Feed the data gods.

If you have any questions, feel free to ask here, or contact me on Steam. As an aside, I'll also most likely be recording the majority of the event for my Youtube Channel!

See you there!
 

InstantMuffin

L2: Junior Member
May 26, 2014
64
48
The results:

blE2UQA.jpg

oN2GcUN.jpg

K4gsT9V.jpg

Vv1Ron2.jpg

qKt8IQA.jpg
 

MystycCheez

L1: Registered
Apr 20, 2013
39
43
This is excellent. Have we had a heatmap test on official maps? If not, I suggest Badlands (Under both pub players and comp).
 

InstantMuffin

L2: Junior Member
May 26, 2014
64
48
I don't know if Valve has these privately (they do for CSGO and actually published them), but someone tried to collect heatmaps for TF2 before my time here, but discontinued.
 

TyeZenneth

L6: Sharp Member
May 31, 2014
340
293
I love the smell of heat maps in the morning.
Smells like victory.
 

henke37

aa
Sep 23, 2011
2,075
515
In conclusion: people tend to die on the control points. Except for the 2nd, that one spills over into the access corridor just behind it.

Also, the heatmap is upside down.
 

YM

LVL100 YM
aa
Dec 5, 2007
7,135
6,056
Conclusions: Either the heatmap plugin is losing all information and blobing everything together, or we picked -THE WORST POSSIBLE- map to test on.

Like seriously, those are horrendous heatmaps.

Can you post the death locations as individual pixels (or really really small dots) muffin so we can see each individual death so we can tell if they actually blob up like that?

(note: being the worst possible map to test a heatmap plugin on does not reflect the quality of the map in anyway or the quality of the heatmap plugin. It reflects how all those deaths have no variance in position and result in a heatmap that tells us nothing)
 

InstantMuffin

L2: Junior Member
May 26, 2014
64
48
Yes I can.
You should know that this is linear falloff. Means when a pixel and its surrounding area has half the kills that the most active area has, it will only have half the temperature.
If you have more data and less smoothing, the regions become more well defined and stuff is more detailed.

Valve's example:
http://blog.counter-strike.net/science/maps.html

You can notice four differences here:

-They use a defined grid, it is not on a per-pixel basis
-There is no smoothing involved
-They have a 6,5 million entry large dataset
-They use a simple gradient between three colors, not four

On a different note: I wonder why you havent spoken up in the last month? This is valuable feedback which would have steered the development differently.

FYI: No information is "lost" in the heatmap plugin.
 

YM

LVL100 YM
aa
Dec 5, 2007
7,135
6,056
Sorry, I don't keep 100% up to date with the forum stuff, I assumed that once a 3hr sesh was planned there wouldn't be anything happening until afterwards.

I don't really like the visual aesthetic of the CSGO ones, but they certainly reflect a lot more precision in the data than their earlier ones which are far closer to what you've made in visual style.

Why are you still displaying blue in areas where there were zero kills?
 

xzzy

aa
Jan 30, 2010
815
531
Questions probably aren't appropriate, he needs to detail what exactly he does expect. It looks like a pretty bog standard heat map to me and there's nothing to complain about with it.

Only deficit is the amount of data backing it, and nothing solves that but months of play time and thousands of deaths. Only at that point can you start to make conclusions about anything.
 

YM

LVL100 YM
aa
Dec 5, 2007
7,135
6,056
Because it is a heatmap.
Zero kills -> cold -> blue.

Maybe I don't fully understand the question.

0 -> No data -> Black -> 0% oppacity
1 -> One death -> Lowest on the scale -> Blue -> Whatever oppacity blue starts at.
30 -> Tons of death -> highest on scale -> Red -> Whatever oppacity red ends at.

Having a defined edge where the is no data vs 1 death will help with the "it's all one smooth blob" effect. Especially for maps with barely any data.

If you still don't understand, like this:
2forts.jpg


Here's a slightly better example, because you can see there were no deaths in the blu setup area, this is important information. With your style, it'd look like there was an even smattering of "barely any" deaths in this area, vs the actual "no deaths"
cp_steel-kill.jpg
 

InstantMuffin

L2: Junior Member
May 26, 2014
64
48
There is no "no data". I can add the option to set a min-percentage of the max score if you like.

EDIT:

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

DgPSekL.jpg

mdpJtJG.jpg

9bfAdzR.jpg

yftYHQ8.jpg

5HFHZhV.jpg
 
Last edited:

wareya

L420: High Member
Jun 17, 2012
493
191
My first suggestion is to define 0 as off/zero-alpha like YM also suggested.

My second suggestion is to do that multi-spectrum gradient thing if you aren't already (I can't tell since you used colors in the order of the rainbow). Here's an (ugly) example (using a spectrogram): http://i.imgur.com/fdi0KB8.png

My third suggestion is to define the window/size/radius/blur of the blobbing effect by how much data has been collected, perhaps for the whole map and perhaps for that general area of data.

My fourth suggestion is to always normalize the output data in terms of amplitude before applying the color gradient effect -- perhaps in terms of bandwidth, perhaps in terms of average deaths over time compensating for playercount. This removes the need to adjusting it manually, and the blobbiness should/will be the identifier for how much data has been collected (so how useful the heatmap is).

Finally, always remember to use linear interpolation and better interpolations in their appropriate places. The color banding effect should absolutely be linear if it isn't (I can't tell due to the map background underlay).
 

YM

LVL100 YM
aa
Dec 5, 2007
7,135
6,056
how is there no "no data"??
I don't understand what you're doing to get that. If there isn't a death in an area, then it should return 0 for that pixel (or area or however you're dividing up space) So in that section there is no data.

You shouldn't be removing areas where there are kills because they're under some arbitrary % value. There should be an incredibly clear "no kills" vs "one or more kills" division. If there isn't, I have no earthly idea what you're doing to that data.