Screeps

« first day (19 days earlier) ← previous day next day → last day (27 days later) »

PhiNotPi

00:54

@TheNumberOne Looks like something is happening in your room.

TheNumberOne

debugging

PhiNotPi

Everything just died.

TheNumberOne

I accidently corrupted the memory of a bunch of creeps.

PhiNotPi

I'm trying to rework my transporter code.

I'm going to try to view it like an electrical circuit, with sources and sinks.

The movement of transporters with energy represents current.

PhiNotPi

01:12

Are there any other rooms that have opened up?

TheNumberOne

My script is working as expected.

I'll work on upgrading my controller tomorrow.

PhiNotPi

I'm launching a voyager into someone else's room.

Ugh... we're pretty much sealed in because of a bunch of other people's walls.

8 hours later…

PhiNotPi

09:43

I got my CPU raised to 300. Now I can actually play the game. :)

1 hour later…

avdg

10:49

congratz :-)

it's not always who pay, who has to win :-)

(my bot dropped to 15th place)

call it 16

1 hour later…

PhiNotPi

12:05

@avdg You are still using the 100 CPU limit right?

avdg

yeap

haha me :-)

PhiNotPi

And yous said it was 2000 lines long?

avdg

thats the code generated code alone

PhiNotPi

How on Earth are you running that within the time limit?

avdg

how did I get the engine not stuck on the same line of code?

most timeouts are still happening from the pathfinder though

PhiNotPi

12:07

How often are you timing out?

avdg

now rarely

got just 1 warning this night (IF I received all timeout warning)

PhiNotPi

I was wondering how a piece of code that long could run so fast, given that mine is 300 lines long and is always over 100.

avdg

though my builders aren't looking nearest refiller atm, they look at the spawn only

while(true) { /* this code is fast! */ }

thats how you get slow code

big decision trees are still faster than search algorithms

and caching helps

avoid recalculations or work that can be done more efficient

(thats why I end up digging in the api to understand how that code works, just to spot slow code)

the only problem with big code is compiling, but the javascript engine got used a lot and got a lot of fine tuning

v8 just converts code quickly, while code is being executed the executed part will be optimized by the engine again

so reusing code would make it go faster on v8

but don't stare blind at reusing code if there is no fundamental to abstract same-ish code

I still prefer functionality, and get merges by cleaning up / generalizing my own code very often

I also have a lot of flag commands, which are never executed unless I add flags

which takes a lot of checks, so it should be lots of fast but long to write code

PhiNotPi

12:26

@avdg If you don't mind, could you help me figure out some ways to speed up my own code?

avdg

how can I review the code?

well, its your code, so you decide :-)

I've put a part of my code public

PhiNotPi

I'll... post it on GitHub or something.

avdg

shhhh?

PhiNotPi

Okay, here you go: github.com/PhiNotPi/MMO-KOTH-AI/blob/master/main.js

No modules are used.

I originally had a module-based design but it somewhat conflicted with my current method.

I also ran the code through a formatting tool, bumping it up to 400 lines.

avdg

Here is some code that measures the difference between time points: github.com/avdg/screeps/blob/master/scripts/_generics.js#L77

meh no worries

I've started with a base

ended up with duplicated code

then moved that to my extensions

now most of my code is in extensions I create new extensions from duplicated code in extensions

or at least, thats how I rolled

github.com/PhiNotPi/MMO-KOTH-AI/blob/master/main.js#L148

I would do it opposite

though it may not behave the same?

inRangeTo support.screeps.com/hc/en-us/articles/…

I'm just scanning the code, not digging in details yet

what does the timers say btw?

PhiNotPi

12:42

The variables aVal, hVal, raVal, rmaVal, etc. are the calculated hits that attacking/healing/rangedAttack/rangedMassAttack/etc. would deal.

If you are talking about console.log("ra" + (timer1) + ", act" + timer2 + ", role" + timer3 + ", upg" + timer4 + ", gath" + timer5) then just ignore the words. They used to be meaningful, most probably aren't anymore.

avdg

my ai spend around 80% of the time guiding creeps

setting up takes about 15%

collecting data and other stuff not much

finding hot spots are ideal in terms of improving the bot

avdg

13:00

also I would use more functions later on but introduce them slowly

it gives v8 more opportunities to optimize code if they are reused a lot

if they are only used once or twice, the code would probably spend most of the time being compiled

under 10 times (? maybe 5, maybe 20 didn't benchmark it) you'll get closer to compile par compared to code compiled once (unless v8 doesn't like to optimize the code)

above 10 uses I guess the code runs faster if reused (by putting them in a function)

avdg

13:27

Hmmm... github.com/…

PhiNotPi

Should I rip my thing off GitHub?

avdg

13:41

I'm not sure, I guess it prefers what you want

personally I don't care as much because if I don't push, code gets outdated quick anyway

(because code tends to improve a lot if worked on it and managed well)

but I do maintain my bot ai code on github itself, but only the core part

2 hours later…

avdg

15:46

ow boy my logs are getting spammed by other bots :p

probably trying to say hi :-)

17:46:11]invadersS ---- 3
[17:46:11]creeps ---- [object Object]
[17:46:11]rooms ---- [object Object]
[17:46:11]

avdg

16:32

that sink system is actually putting more load on the pathfinder

just noticing that when I'm generalizing the system to my transporters

already aiming to get such stuff done for few days, but transporters are complicated and I had to write a lot of code (and test it)

PhiNotPi

What sink system?

avdg

its actually more supply and demand

bringing them closer together

it seems that when both are trying to find the closest, it may seem to use a lot of time

but I need to box the situation before I can make a good conclusion

its actually a double sink system

and the target is a moving one

hmm maybe that is the problem

uh, I need to fine tune the path cache I guess...

4 hours later…

avdg

20:24

lol, my guards are protecting the edge

3 hours later…

avdg

22:57

Need to benchmark this function I guess

https://gist.github.com/avdg/d4dfe33cfd858cdcf58a

« first day (19 days earlier) ← previous day next day → last day (27 days later) »

all rooms

Transcript for

May23

May '1524

May25

Strategy discussion for the Screeps programming game

join this room
about this room

00:00

06:00

12:00

18:00

all times are UTC