next »

[JLJac]

Hahahaha if you guys didn't notice the blinking heads until I mentioned them, maybe they're not that big of an aesthetical disaster after all  


I agree that something has gone missing in the smoothness of the new build though. Research will be directed at finding out exactly what!

Update 305
Dead end mapper! I think I've managed to identify dead ends pretty well today, stay tuned for a million screenshots so you can judge for yourself!

When generating these I disabled short cuts to make it "what you see is what you get" - there're no secret passages between tiles, so what looks like a dead end in the image is actually a dead end. If this is confusing to you let me know and I'll try to explain better. Some images have something that looks like a very obvious dead end but is greener than anything else - those are room exits.

Essentially the way to read these pictures is that red is considered a dead end and green is considered a good area to be in in regards to not being cornered. The AI behavior I have in mind will be able to use these maps for a few things, but I guess the gist of it should be that if you're being chased, gravitate towards greener areas rather than running into the red. If your tile is redder than the tile of the threat, you're probably being cornered and it would be a good idea to find a path back to greener pastures.






Same level mapped for Fly and Slugcat respectively. The fly obviously has access to air tiles, while the slugcat is stuck on the ground.






See the pink tiles? Those are "fleeing targets", essentially tiles that are A - in "hub" areas from where you can get to other areas without being cornered and B - somewhat evenly spaced. These are few and well defined for a reason - when a fleeing AI doesn't know where to go, it is supposed to be able to evaluate those one by one. Which one of them is closest to me? Which one is the furthest from the threat? Which one of them can I path to without the path intersecting with my threat?

In the difficult scenario of knowing where you want to get away from, but not where you're going, these fleeing targets might give a clue. If you arrive at your fleeing target and still feel panicked you can just pick another one and run around between them at random - you're at least guaranteed not to end up in a dead end.


Here's a little comparison - the same map, mapped for the fly creature, with shortcuts turned on and off:



The rooms on the left each have a shortcut in them. Note how when the shortcuts are turned off these rooms are pretty much completely considered dead ends, no self-preserving creature would allow itself to get chased in there. When the shortcuts are on, however, they get a path through them, making them more considerable as fleeing destinations.

Last for today, a thing that I'm not keeping but which ended up looking pretty cool - I tried to have the algorithm grow at random, and the effect ended up looking like a tesla coil, or veins:



Edit: Just came to think of it, room exits should obviously have a fleeing target each as well.

[jamesprimate]

THIS. UPDATE. IS. AWESOME.

[JLJac]

Ok, so here's a technical question.

These maps take too long to calculate, so I'll just do it once on the first time the level is loaded, the write it back to the level file and have it hang out there.

When I calculate these maps a by-product of that process is a dijkstra map for each exit in the room as calculated for each creature. This is really useful info to the AI, it could do a lot more and a lot faster if it had access to it. Sooo, I'm thinking about perhaps storing that as well, when I'm at it?

Say that a standard level is 50*40 = 2 000 tiles.

Say that I end up with worst case 15 creature types that need to do these calculations. Some creatures share maps as they have the same basic movement, so it'll probably not be that much.

Say that a room has 4 exits, all of which I'm saving a map for. Then I'm also saving the dead end map, so 5 maps all in all.

2000 * 15 * 5 = 150 000

So, a hundred and fifty thousand integers. If the level is bigger, or there are more exits, that number could be even more crazy. If the level is 100*80 instead, we end up at 600 000 integers.

The level text file I have currently is at something like 4kb, and contains something like 3000 integers to store the level geometry. So by that estimation a 150 000 integer level would end up at... 200kb. That doesn't seem too bad? Modern computers have gigabytes of RAM, they should be able to hold those amounts of data? The bigger level ends up at about a meg.

What do you guys think? Is this viable? It seems like such massive numbers. It also kind of is a waste, as not all creatures are in all levels. Also, if you look at the screenshots above you'll notice how in the most cases the majority of tiles are not accessible, meaning that a majority of these numbers would just be zeroes.

When loading a level its text content is held in memory, then converted to rain world code objects. Then that text is discarded. Is that a huge hit for garbage collection? When the game starts up each level is loaded once to retrieve some data from it, and if there are hundreds of levels I worry GC could get overwhelmed.

Is having all those integers in memory during play a performance problem in itself?

Thankful for all advice!

[JLJac]

Just struck me that I could do a very easy compression of the maps by just going tile by tile and only saving them if they are indeed mapped, leaving out the blacked out terrain. Then when I write the data back, I just go by the same order. It seems from my tests that this would get the amount of integers down well below 30% in the level layouts I have going now.

I could probably do with dijkstra maps that are just tile-by-tile rather than creature specific as well - not AS good, but still an asset to the AI. Then I would not have 15 * 5 maps to save, but just 15 + 5 = 20.

(15 + 5) * 2000 * 30% = 12 000 integers

That shouldn't be a major problem, right?

[JLJac]

Update 306
Well, most of the day was spent in hesitation over above issues - I did some calculating back and forth but that obviously didn't help me much as I do realize that what I think of as "big numbers" or "small numbers" is irrelevant to the scope a computer. I didn't want to start implementing it without knowing it was going to be an actual realistic solution, so I kept away from it and did some other poking around today.

For example I made the dead end mapper use the creatures' terrain preferences rather than a first-best grab of tiles, meaning that now the maps will hopefully gravitate a bit towards terrain that the creature is comfortable with.

Another thing I did which has been a long time coming was implementing a terrain proximity map in the common AI map:



Any flying creature will be very thankful to have this info, as it allows it to know what's open air and what's uncomfortable tight spaces. Also, if we want to add a creature bigger than 1 tile wide, this allows that creature to know where it can move.



Here's a dead end map for the fly with a preference to stay away from walls, looks pretty reasonable for a fleeing behavior for a flying creature right? Given that those bottom two corridors are room exits, otherwise it shouldn't have anything to do down there.

[Sebioff]

Sounds like massive numbers, but that's really nothing
GC is problematic if you produce garbage during gameplay (remember that not all memory allocations produce garbage), as every once in a while the GC would become active, potentially resulting in a noticeable FPS drop. It shouldn't be an issue on level load.

Nice data maps :D

[gimymblert]

Yeah the way those numbers lay out in memory will be the biggest "problem" and even that won't seems to be a problem at this scale.

However never forget path finding happen at forking tiles (more than 3 link) the rest can be simplified as waypoint (2 way path between fork) or area (box area). Those can be achieve through pre process or at runtime by evaluating the type of tile and skipping unnecessary evaluation.

For exemple jump point a* use this process to accelerate greatly grid base pathfinding by looking box ahead instead of evaluating all the tile. http://gamedevelopment.tutsplus.com/tutorials/how-to-speed-up-a-pathfinding-with-the-jump-point-search-algorithm--gamedev-5818

[absenter]

I had a chance to play this game briefly at BostonFIG. It looks as incredible in person as the GIFs we've been seeing. I must admit, I felt like I was terrible at playing it. I'm excited to spend more time playing it when it comes out and getting a chance to learn to play it better! Very cool!

[Thomas Finch]

That looks so awesome!

[jamesprimate]

I had a chance to play this game briefly at BostonFIG. It looks as incredible in person as the GIFs we've been seeing. I must admit, I felt like I was terrible at playing it. I'm excited to spend more time playing it when it comes out and getting a chance to learn to play it better! Very cool!

yo! was nice meeting you! thanks for stopping by and hanging out

TIG was pretty well represented at BostonFIG. i think i met more forum people there then all PAXs combined. i guess that makes sense with the indie game focus, but still seems surprising considering how much smaller it is.

on a similar note, the RW booth was NUTS. we had a crowd for 8hrs straight doing constant 4 player matches. Its possible we had more people playing it in one day at BFIG than 4 days at PAX Prime. and that was with the most slap-dash, ghetto "booth" setup possible 

[Christian]

So any more impressions of the game so we can experience the awesomeness vicariously?

[JLJac]

Hi!

I usually don't work in the weekend, as I think a five day week is probably the better option in the long run, but this time I made an exception and did some polishing on the dead end mapper. Now it works like I originally intended it to - actually finding loops, and from every loop drawing a connection to every exit. Tonight when I do my actual update I'll throw an image of the slightly improved dead end mapper in there.

@ Sebioff, Jimbert, thanks! Just needed a reality check. I might as well implement those solutions then, if something like a million integers isn't necessarily a problem. I have been looking into Sparse Arrays some, as I think I could use something like that for the heat maps where the great majority of the level is just blacked out, but from how I understand it there's no way to get the smaller memory size without trading it for a more complex look-up process. And you guys have told me that if I'm in a situation where I have to choose between burdening the processor or the memory, I should go with the memory - so in that case it seems unnecessary to go out of my way to go against that advice.

I had a chance to play this game briefly at BostonFIG. It looks as incredible in person as the GIFs we've been seeing. I must admit, I felt like I was terrible at playing it. I'm excited to spend more time playing it when it comes out and getting a chance to learn to play it better! Very cool!

Thanks for dropping by! Sorry I couldn't make it this time haha! One fairly common complaint with the game that I remember from PAX was that the controls were difficult... So you're probably not terrible, it's just that the controls aren't as streamlined as they should yet. Some people learnt the controls intuitively while some where struggling with them, so we probably need to find some kind of common ground between all users here. I for one have been playing with these exact controls for three years now, so I'm way too used to them to be able to evaluate them properly - I think that when we get down to streamlining the user experience we'll have to bring in a few people who haven't played the game before and watch them play to see what problems they encounter. Anyhow, glad you liked it!

This curve:



is there a way to get a curve like this that fits snugly in a 1*1 square? I want 0<X<1 and 0<Y<1, and I want f(0) = 0 and f(1) = 1, but no X other than 0 should produce an Y of zero, and the same with 1. In english, the graph should hit the corners of the square perfectly. The one pictured actually doesn't hit the corners, if you change the steepness of it the graph will break out of the square elsewhere.

Edit: Yay! http://dinodini.wordpress.com/2010/04/05/normalized-tunable-sigmoid-functions/
It actually really isn't a sigmoid I think, as it doesn't have e^-x, but it fulfills the same purpose.

[gimymblert]

Use unity's custom curve or use a bezier or equivalent!
http://docs.unity3d.com/ScriptReference/AnimationCurve.html

Or maybe itween or gokit or hotween (all free with source code I think)
https://www.assetstore.unity3d.com/en/#!/content/84
https://www.assetstore.unity3d.com/en/#!/content/3663
https://www.assetstore.unity3d.com/en/#!/content/4562

I think what you are looking for is called "ease in/ease out"
http://stackoverflow.com/questions/3367308/math-ease-in-ease-out-a-displacement-using-hermite-curve-with-time-constraint
Here is an answer about but that's basically advice 1, you would remake unity's animation Curve

[JLJac]

Update 307
I spent the day trying to accomplish serialization of a class, as that was going to be a time saver I believed  Serialization is basically that an object is converted to text (which can be stored in a text file, for example) and then converted back to an object again. My idea here was that I was going to save the AI maps like that. After having broken my game pretty badly it became more about trying to patch it back together again, something which was made harder by not being able to make visual studio break at exceptions, and... Yeah you get it!

The end result was that I actually managed to make the serializer spit out some text, but way less than what could possibly contain the data I was trying to store (just the names of the classes I was serializing, basically. And some.... web url!? Idk.) To serialize an object I also have to adapt in a lot of wonky ways, like having an empty constructor etc. Also, all objects that are stored in the serializable object also have to be serializable... Basically, at the end of the day, I think it's going to be more efficient for me to just make my own method for converting my data to text and back again. Also, that way I will positively know that it's not storing something I didn't ask it to, such as strange web URLs and the like.

The only problem I can see is that I need to store some floats. Integers will become 1, 2, 3 or 4 ascii characters, which is kind of wasteful with space but still within the realm of the reasonable. For floats however, if I want to save them with somewhat good granularity, I have to use like 6 characters for each, which seems horribly wasteful compared to the fact that a float is actually no bigger than an int in raw data.

Does anyone know of any cool ways of converting ints and floats to ascii text that isn't horribly wasteful? Maybe you could make a huge BitArray, feed it all the ints or floats or whatever, then cut it up in pieces of however many bits make up an ascii character and just put that character in the text file? That would surely be using the space to 100%, right?

A pretty frustrating day. It never cheers me up when what I'm fighting with isn't an actual logical problem, but the functionality created by the microsoft people - all due respect though, everything they've made that's streamlined and easy to use I obviously never even notice. The day wasn't all bad though. I did get some pretty nice restructuring done when it comes to "preprocessing" levels, ie finding out AI information about them and saving that to the text files. Also I solved the curve problem of this morning, and made a curve visualization tool that I'm going to have a lot of use for in the future:



Over and out!

[JLJac]

Hm, I've managed to translate float arrays to characters bit by bit... It doesn't become ascii though, but unicode.

Look at this insanity:

饕䲟㌵䮈綷䲫᠄䲇휕䰍ൣ䱷䚼䲘⧓䯅蠄䬌೎䨷쫿䲆쇿䥙ꌂ䲫騨䱌䱃⦏䮌萕䱀떵䱹ꮛ䲶쬩䪎䮲海䲝땘䰠歐䱌䯒︇䲺䱽퍜䱊䲧⿹䬚躨䪋﾿䦬묄䬵䕫䲦��䰣궧䰍ꮋ䯭诺䨞憎䱑₏䱬䑈䱱璚䱞怒䬺欤䯿㉐䱽擀䲑㾇䲛ᷩ䲭诞䮈뜐䬆삇䱍졣䨆宴䦏횛䰹㑅䱙槥䱩筏䮜ꠢ䦉턚䲫敂䰥��䱄힩䦪ꏈ䲮ݺ䲠⒰䱠䃓䲎곀䰘옶䱲眶䫐ꌩ䰾撦䲟뼮䩏寧䨪亘䮘몗䱦Ꮺ䲟肎䲺곉䲶츴䰰믋䱘孃䲕硰䲹��䲋綹䱵䭌烋䰞쬖䲕㐹䱸䰹ષ䲹倣䲲ŭ䮛祂䲒弣䰜㶊䱙ᘤ䬂綍䱚⽁䰡伵䲫⯐䲋㩭䮼᢯䲼䀬䱩䲆竡䮏ﵽ䰱茚䯵谏䰼栏䲚ས䱸ẗ䱴켁䲟⊍䱞俴䰂쨩䲉⛎䲃秅䲼䲙슉䱱괹䪃䯎ᗆ䯇䯉籣䮇䱴伖䱠䱗ᒷ䛤툀䫸騔䮏ꨦ䰣뀫䱀仆䲌뷈䱕ꆳ䭈죹䱶፦䲛姈䰾⊽䲀ꚟ䢣塟䲔궼䱨禷䭝憣䱕藩䲜휹䮍爳䧚䰭蒺䲏︥䯧帚䱥캃䲏㊢䲺友䱁쒴䯹ꬾ䲸䧯ᣄ䬠袭䲻蝽䯩ㇽ䮷翛䨇㒃䯸켬䲑ᥫ䲝䯝腗䪔䓋䲋腐䮕ꋪ䰲覯䫍ԍ䰃졄䰒풃䲔죝䫼䊗䲂뵘䲍䲻덖䰼바䫀✞䲜釶䰯낈䮞䯏풺䱡䁶䲸䱜ﶨ䲡溴䧜峏䮤曬䯞䂷䱼ꁹ䲽똵䰤儈䯟씦䲃偟䯽䰓䮏䦛롩䲶㌔䲪⤃䱫쉔䲢嶮䲍컓䱌䬳렺䱸ᮍ䰍䪴䧛鿰䲄䯀팗䲢먙䲧侓䰠䰹㳔䰎櫯䲨㺜䲮䜣䯹䲷嫪䯵럥䲢䬱鏉䰯瞂䯒ႇ䱁ꑥ䲪㐤䮚罼䲢픚䱏㸶䲪꧴䰸馉䮬ᕻ䲶곟䲟ꄖ䯌��䲮餇䭞㌮䰶㫹䰢蒠䯠䰈譠䪓揤䮴蠪䲬壧䲴铅䩿ꁆ䲟晖䱁䲥窧䲾䨙䱩枊䰪❹䰄䰮蘈䯜��䲯稘䨌讅䨈쁢䰽癚䫏侌䯯ɴ䲘⋦䯴丐䲆ᫌ䭑蟒䲠圲䰔䋤䱫ꖅ䰋绑䲗题䮧샂䲵崤䩹凊䮌᤻䲦䫨䱓슷䲞Ⓞ䲐ᱠ䯵릥䥖筫䲐ꂟ䮘ᤏ䲄⟩䱡䷬䱏䰄䱚燜䲮ᔏ䩧襵䱺᳤䯚턳䰂⯤䱡푼䱲"䱆쿙䱈䫐䲅䲏ಪ䬞曋䲚軈䲩ৎ䱲䯵䰣⤆䲳襁䱦懦䨺春䯡ꮙ䮶꼟䰕국䭮곇䲲搣䱖��䰮윋䱵⻧䲟ᑱ䯈蓩䯄䱄僬䫋܂䱫俾䰅鏅䮅慐䰊䱵爿䲐妝䱞眄䰉춭䲰얜䰒ϰ䲉䭁抎䲼��䲴��䲈邷䬐蝔䮀䮰䲹趽䭂ᐳ䰁㥨䲁䮑჈䫾➾䰫顖䰧㨤䰔냒䫅䱜㺲䧵鬣䲼뤴䰢䬱辢䱍ꗅ䰱䟃䪊窤䲓䲝㨆䲚䓋䲈홵䭣쮹䰰ꦁ䫇燻䱳쎝䮩ﺸ䲸䲱��䲻䵠䱗ܽ䯫잔䰖꿶䯕䰉ึ䬈ߐ䱈섓䮇ꡝ䭱楢䲞헻䲘ẁ䲤㔔䬅㨂䣸ᵆ䰈東䲸臢䱅ꭹ䲞ᥤ䮳彡䱪��䬿砠䱬맏䱕ງ䲬バ䰻㤧䰺⠣䲱঻䲶᜜䬘锫䲠鵥䱃ፓ䭫鲫䲟Ⳅ䲵瀤䱴柊䩤ꢧ䯕鈫䯉��䯑僀䰵籏䬔��䰡僤䱊Ꝗ䨂贆䰚ᇬ䲋菌䨨嗠䯆봅䭹喊䮧鶍䱵纤䮁侚䰫橌䯄㕹䱝ᣏ䭦ퟵ䰟塣䭇谂䰖䯮ʑ䮵晸䲖��䱰��䲄ꮰ䰫勍䲋䱿갳䧫蹯䰳℄䮝꨷䯋䕿䬄陥䦫慢䧥ஊ䱡⦗䭠䓑䲾䯴￥䲠︧䲈뙴䲫歔䩶﫿䲻趼䲓쇖䯇䄏䲙敧䯱譅䰥賡䮹枂䪅ﵽ䮤䯽쒠䫚撔䮄��䮴䲥䰁빵䫐춊䫽邴䪂㮧䰡孂䫦��䯋ꈙ䰻䯗䯱䨳䱍쟇䩞鋰䲉䰦轟䰛퇗䲪钧䮘䘰䭃轋䮈﷭䰿ႅ䲀捑䲇䔏䨥芶䮷Ჸ䯷㵐䰇엞䲆䟸䯲シ䲍ᧂ䯟檣䲀뮀䱃搰䱝⇿䱋縎䮝姁䲸ꬕ䰫廓䱱ꓩ䱽퉖䰈寢䰃㇧䲄艬䱝䍰䱍쥪䲱茢䲲㔪䱀놿䬁僅䲥〴䲺醰䭆캆䲤栃䰟䲳ﭑ䰞귚䪯漉䯳ǰ䲍宠䲥˰䲌윱䫘᦯䭘䀚䰩鎥䭨W䮺ㄌ䰁廒䰕ᆀ䱏伆䱷⢀䮼岣䱩䬠錽䱧♷䱳爥䲐ᮣ䱌즐䰜誅䯍ｑ䱹戼䲤䨆䥟䰰좿䲅቉䲖䧑䲧ࢁ䱁憄䯱꞉䱥㞕䨽ꭘ䲨巌䲪끵䬨ꀤ䰆餺䱌��䱠賦䧦㝴䰱ᘪ䲐଒䪐㋯䲘岫䲔䦚䲪৤䲖达䲦揃䲈胖䰝缘䩮硺䰋ꅮ䱉䮽䲜䯞䲒젨䦨ꚕ䰗��䣸纶䮁듐䮄뤦䱥Ⱇ䲇씅䮯땢䰽፠䱾旪䯆ȣ䲟渟䲠࿎䲑懚䲢ݑ䯱曠䨡ೡ䰁訥䪀噀䪚罻䲦颞䭣꠨䲨⏰䱢닸䡥쮷䰕츼䲨粈䲅쪶䱀嬎䭵鴀䬶ꌚ䲩쨱䬜霚䲆ѵ䱦翺䲘䲠嫎䱐䰟媲䮬䥲䱑䲲鷋䲤��䲓胶䲒쀁䯃崩䬛苩䯝鐍䲗蓢䬒搩䲇䫼䱈嚢䰴소䨟䲮栿䲥▓䭽陔䭷萐䲽ず䩵ନ䲓ꨆ䦼㝀䰦铚䲢೫䰒��䲳⃘䲱쮅䲺挤䱥邕䲙뫤䲊敊䯨쥤䲭Ἁ䰫咮䮯ｍ䲖렝䲞䨃䱒ᇞ䱰뫈䰓똥䲡䱱㪷䥤앋䲭ߴ䰾卅䲞텷䲯ฤ䪽ꎈ䮱쪹䤸ȿ䮙䜶䬡ඟ䲛㦜䯎츧䮾䰢䬞㏰䯵怂䰳顥䰀炭䱉ꂲ䰍Ɠ䮘졫䮧疪䲕梾䲎꒏䱁ᠵ䲍䯻ޞ䲜इ䬹��䪯꣨䱪闫䲭휵䲲쳣䰴ḣ䲢轶䯷均䲙ꐦ䲴茕䱞厺䱯덴䮇ꯑ䰢ٓ䰩䧞䰏彖䪘䬑뮲䲒ᛡ䲫᠝䲙摣䲯뮇䰬莋䯂ᄇ䱰ꓤ䮋졮䮚璋䮆팣䲟퉀䲸嬊䲄פֿ䲹獳䲨軭䲩䨘䯟㯻䲌⾴䬄Ჶ䲱圥䮧纆䲷㒈䱨헐䱇ᎊ䱠슚䲺㙺䰒֝䮶焨䮔︋䲍읝䰏䰽ﳡ䰮餩䰭䯯慢䫥␺䲯ᱢ䱔✉䱲↪䲘蕝䬧㼚䮾摩䲺鲌䲑䱺崽䲸竐䲐쁅䲀꽦䫖㺳䰼Ꟃ䲘隟䱐姕䲕皢䮛䀓䰯鮘䫜◧䫽��䲣⍌䯐魦䰒䞍䲺詿䲖訨䯬茋䯗奷䯌૸䩪좢䦶롽䮗䲍ⓧ䲻ᮞ䰱舖䰁ὥ䲱緵䲹䱒䣱䫬좈䢄鳪䯰ꃭ䱋狊䬓씳䭶㥆䱸₹䱍斧䰷礏䯍䭪��䰡굴䱛術䱦嵯䰃逢䱼淘䲦翍䲬ῒ䰉Ά䰃䲷ퟆ䱞ጓ䮉椱䲼ꇣ䲐䮐ꖓ䲫览䯭凈䰮糏䲷猼䮚쭝䲫䲀㌲䫁劙䲣魟䰥㍲䲪ꀫ䱘Ⓦ䬚䲔ز䬜쇌䰋䲙欅䲦떥䲚ᚷ䯜狇䲼恽䦾ꆭ䲏䲡殾䬲ꃢ䱧ዏ䮌뱅䲔诺䲞棣䲴䨥鉺䯩噑䱛գ䭸箬䱠苖䲳ᴃ䲊☥䱣ꖆ䭘䢮䯝ꥂ䲴⥾䪥䫥ᣞ䮥␌䯏臏䮉ꢙ䲛㉍䮴뢳䯒ᤱ䲂㫋䲈䮩躯䮾썱䧩햇䰃䩪㸭䮑쟊䭼莖䰊崣䲃娫䬨ꈉ䲽ᜲ䲟㫏䰖汕䫚ꚯ䲐둥䱰掍䲕쑭䲨ൊ

If nothing else it's a somewhat interesting display of what languages have a lot of signs hahaha! This is generated from random floats. The vast majority seems to be Chinese and Japanese, which makes sense. There's quite a bit of Korean in there as well, but as Korean is a phonetic alphabet (although you group the letters together in compounds of 2-4) it's a lot less. Soooooometimes you see a roman or Cyrillic character in there.

Huh. Pretty cool. Just not sure if I can actually store my data like this. It does seem to be re-convertible, if I ask for random floats back from the string-to-float method they're aligned. But it doesn't feel very safe if this game is supposed to be compatible with many different platforms. Also I read somewhere that some computers arrange their bits in the opposite direction, in which case the data will be totally jumbled.

I should probably just do the actual serialization haha 

[gimymblert]

If you want to pack number into less character what you are looking for is not text files but binaries. The question is, does it have to be human readable?
and memory should not be a concern I think. You are over thinking it.

[JLJac]

Update 308
There was a thing called Convert.ToBase64String which makes bit arrays into strings that are (from how I understand it) dense information-wise, and using that I finally managed to save my AI maps to the level files and load them again. These strings keep to ASCII characters and look a lot less menacing.

I have also created infra structure for checking that all levels are up to date upon starting the game, and if not doing the AI map calculation on them and saving the results to the level file. When the level is then loaded it'll unpack and use that info. This part took a bit of fiddling, as when the level is loaded to generate the AI map it shouldn't load the (non-existent or out-of-date) AI map, and when it's later loaded for real it should wait to apply the AI map, as it needs to generate some parameters from the level geometry before it can apply the AI data.

So yesterday's problems are solved! Instead a few others, more harmless ones have popped up. The dijkstra maps don't look perfect and probably need some messing with, and the thing that maps the distance between every exit and every other exit isn't plugged in and tried out yet. But the basics of pre-generated AI data is in! Yay!

In less happy news James told me that Gimym JIMBERT is apparently taking a pause from tigsource! D: This means that Project Rain World will have to do without its main developer for a while, I will however do my best to step up and fill the position myself. Thank you a million times for all the useful help, I wouldn't be half as far without it! Come back soon! 
 

[Cranky]

You shouldn't think about serialization stuff as wasteful, its pretty normal nowadays to save everything in wasteful formats like xml, its fine.

About the curve stuff, its called easing, I just recently implemented that stuff myself, a couple of resources you might find interesting:
The ease-in/ease-out curve is called smoothstep, very straight forward to implement
http://en.wikipedia.org/wiki/Smoothstep
Some other function based easing:
https://github.com/jesusgollonet/ofpennereasing

Most animation software uses bezier curves for easing like here:
cubic-bezier.com/
Implementing bezier curves is rather straight-forward and easy, but traversing a bezier curve in fixed time steps is non trivial and involves a lot of math.
Here is a very helpful link that explains everything in detail:
http://www.niksula.hut.fi/~hkankaan/Homepages/bezierfast.html
It makes traversing bezier curves very fast, so you can use bezier curves for any kind of easing.

[gimymblert]

In less happy news James told me that Gimym JIMBERT is apparently taking a pause from tigsource! D: This means that Project Rain World will have to do without its main developer for a while, I will however do my best to step up and fill the position myself. Thank you a million times for all the useful help, I wouldn't be half as far without it! Come back soon! 
 

Damn it!
I Said I'm taking a break from OUTSIDE TIGS to focus on my game (which is reported INSIDE TIGS), I reported this news inside teh sjw sub forums

[jamesprimate]

deep sighs of relief were heard all across the globe!