buffer[]'s posterous

Speed skin loader

I'm sure many players got many skins, but there is a main problem with that : Teeworlds hang for seconds when it loads all those skins.

As I got myself 226 skins (:D), I've made for a while a special skin loader.

It's simple : it don't load skins until they are used in the game, instead of loading them all.

In game, you'll only see a weird thing before seeing the skin for what 0.5 second, and when you got in the skin selector menu, you'll only see loaded skins, other will be nice weird thing :D

With it, 8240 ms become 1794 ms, with those same 226 skins, so I think it worth it :D

Download it (exe and patch)

A few patches

I was thinking on few things which could be improved for Teeworlds.

The editor is being reworked, so I'm thinking a compressed way for saving maps, and correct one thing (reload the map only if I'm rcon authed and if this map is being played).

The optimal window size should be set instead of 800x600 on Teeworlds first launch, many people won't change it, and it look creepy :)

And I'm actually trying to improve the autocompletion : first, client's and server's commands should be separate (for now 'broadcast' or 'kick' can be autocompleted in F1). But the main thing is, when you connect to a server, all server's commands and variables are sent, so they can be autocompleted in the client. You'll say that's useless for standard servers, but if you're the admin of a modded server, with many commands, it takes sense.

Anyway, if you have other ideas, suggestions or comments, feel free.

Flow

Since I saw those videos, I was interrested in remaking something similar.

After some research and work in tthe Teeworlds code, I've done something working.

That's not finished, I haven't done nice grenades explosions as they were, and the flow thing isn't really finished.

That's why I need some feedbacks, so please take a look and answer here.

You'll need this teeworlds_flow (linux version is on it's way, just need to reboot and I'll post it).

N.G.T.

Just to present the project I'm working on :)

Starting by the start, N.G.T. stands for Neural Group Tees.

Those are server-side bots (as ZBots), for TDM or CTF (vanilla, N.G.T. vs others).

Tees speak for itself :D

Group, so they'll work as a team, I'll work on team-job and work distribution.

And Neural, cause they'll be neuron-thinking.

What the hell is neuron-thinking ?!

Neuron AI is a type of AI which try to reproduce the human brain in its way to make a decision.

In a human brain, neurons are inter-connected by synapses, and pass information by eletricity. I'll take an example : one neuron is supposed to make a desicion, and must take two parameters for this (I'll call them param1 & param2). So this neuron is connected to two neurons by two synapses, which pass those two parameters. But imagine param1 is more important than param2 : for this to be considered by the decision-making neuron, the synapse's size (the one which transmit the information) vary. The bigger the synapse is, the more important the information is.

In programmation, the neuron which make the decision (output either 1 or 0), sums up all incoming values (parameters), but to consider their importance, it multiply them by their weights (synapses's sizes). Then, it compares the sum to an activation value, and output 1 if the value is greater than the activation, otherwise 0.

Ok, it makes a decision. But why is it this interresting, that could be done by some if and else, couldn't it ?

That's where it become interesting : the trick is "training"; You can change weights, so it'll change the decision.

There are several techniques; You can before starting make them make (nice) decisions which you already know the solution. So if it fails, you change some weights responsables. You can also make the weights vary at each generations, and take the bests ones for the next generation.

I choose the second, taking for generation spawns, and best by a combination of kills, flags, time, etc.

Oh and btw, Lanta is doing a similar project for school, they'll be called LBots.

Looking forwards a nice clanwar :D