DDraceNetwork - developer

we just can find analog of letters

00:02

in "

"

= O

= X

00:02

etc

00:03

and so we can create database of these analogs

we have a database of those already, see confusables.c / confusables_data.h

06:10

Created from the official Unicode document: https://www.unicode.org/Public/security/12.0.0/confusables.txt

06:11

Someone just needs to implement a str_utf8_find_confusable and then you can use that to search

def-

[ddnet/ddnet] Issue opened: #1780 SQL server crash in executeSql

``


Core was generated by

./DDRace64-Server_sql -f servers/8305.cfg'. Program terminated with signal SIGSEGV, Segmentation fault. #0 CSqlServer::executeSql (this=this@entry=0x0, pCommand=pCommand@entry=0x55ecdd83379f "unlock tables;") at /home/teeworlds/src/master/src/engine/server/sql_server.cpp:207 207 m_pStatement->execute(pCommand); [Current thread is 1 (Thread 0x7fbf36d3d700 (LWP 20325))] (gdb) bt #0 CSqlServer::executeSql (this=this@entry=0x0, pCommand=pCommand@entry=0x5...

I think we had a utf8 function that reduces strings to basic characters

@deen the comment was adressed 5 days ago https://github.com/ddnet/ddnet/pull/1777

Prefer downloadedmaps over maps ( #1702 ) by ChillerDragon · Pull...

Fixes corrupted demos if there is a different version of the current map in the maps/ folder. Thanks to @archimede67 for initial research. Who ever implemented that probably thought maps/ is a good...

09:43

imo the additional crc check doesnt belong in this hotfix. It is heavily tested and fixes almost all bugs demos have currenlty. The longer we wait to fix this crucial bug the more demos we loose.

alright

def-

[ddnet/ddnet] Pull request closed: #1733 Store colors unsigned

10:26

def-

[ddnet:master] 3 new commits

0d39eb1 Use unsigned integers for colors - Learath2 db5c17c Avoid implementation-defined behaviour - Learath2 bbb1b2c Merge pull request #1733 from Learath2/dd_pr_colorsigned - def-

Hey, anyone up for some DM? I started a project where I collect gameplay data, my goal is to use machine learning to teach an AI to play teeworlds deathmatch. My first approach will be a deep network. Broadly speaking I render a highly simplified scene of what the player sees at a resolution of 90x50 and record the player input at 25fps. Join timakro.de:8500! I will release the source and collected data in the future.

3

nice!

11:06

but i still cant join the srv :c

Wow, thats awesome

@timakro Nice. Is it learning from the players?

12:34

Can you make it learn from teehistorian files?

@deen The plan is for it to learn from players, I've done nothing on that end yet. I'm not anywhere close having enough data. I think ddrace would be too difficult to start with. Does teehistorian include the environment data (could an image like the above be reproduced with teehistorian files) or just player input? (edited)

I figured that you need more data. We have hundreds of GBs of data, but only DDrace

12:48

just player input, but the map file is linked of course

12:48

so with map file + teehistorian you could get something like above

Yeah, I don't think I could possibly do ddrace, it's incredibly complex

Just give it more neurons and rent some cloud gpus?

12:50

But yeah, solo maps would be easier to start with

It's a lot of work even thinking about input and output and preparing it for the network

are you using a modified tw client?

All I've done yet is write some server code to save compressed gameplay sessions

12:52

In this format which is relatively straight forward to feed to a CNN

"compressed gameplay session" is exactly what teehistorian is

Why whould he put "Oxygen" to the second nickname?

13:10

@justabutterflier >for example, player with unicoded-nickname OXYGEN (just example) will put to this line (second nickname) "Oxygen" and other players can find him with nickname Oxygen, but on the server he will be an OXYGEN (unicoded)

@timakro maybe @noby is interested in your project. He has the most visited fng server.

https://youtu.be/8ocxnWipQDw

timakro

Teeworlds Machine Learning Project (TMLP) - Training data example

def-

[ddnet/ddnet] Pull request closed: #1778 respawn is now instant, but rate limited to 3 sec unless player clicks mouse1