itch.io

As I said in the new forum, this can be closed if you want. I reported the data in the wiki already.

Well I see them like "ok, leve up, more variety and more difficulty"

I see that you are laerning how to make the forum aesthetically awesome.

I'm not sure that there could be a better system actually. The problem is (a) the map and (b) the fact that players learn.

Nice subdivisions by the way. phpBB for the win.

There will be always someone that will complain (proof: find reddit submission with at least 100 score and only 100% upvoted - impossible. Quite a statistical proof). The point is, does it helps to: (a) use all bots, (b) make fair conditions, (c) satisfy you, (d) etc.

For me it is fine already.

Well is it not good? 1500-1700 is the top, You cannot gain so much more.

I do not know if a subtree is counted or not to determine the priority of a branch of the tree. Like: place it wherever, it does not matter.

This may get a sticky post otherwise people will write here.

I was reflecting over https://itch.io/t/47112/so-how-are-we-supposed-to-...

A way to solve this would be (and would be also raise the variety of games a lot): the actual system decide for the players the map (it is not that the player can decide the map anymore), but the players can choose type of bots and decision trees.
What if the system decides the map and the type of bots? (since the bots are simmetrically places, there should be no advantage for any player).

For example split teams: the system decides (in one instance, every instance is different) that each player will have shotgun, assault, machine gun, assault, assault (from the north to the south). So both players will know in advance which types of bots are on the map, and then the challenge is to select the proper AI for this configuration.

Surely for the ones that wants control this is not so great, but it will make the focus only on the decision trees. It would also solve the problem that the assault class is not used, since the distribution of the classes will be random. And this means not so much change in AI development from a version to another.

So a player would be neededto develop a generic AI, instead of having none, and then optimizing for bot classes according to the map and mode.

This forum has serious problems to accept input (firefox). Now I start to write in direct html. I see no problems with the raw html edit

My question is: sometimes I have very long connections between elements and I may forget that one connection is shorter than another, so the priority changes. My idea was: can I use subtrees immediately near to the root node, to use them to fix the priority (as far as I understood the priority is determined by the position of the end connection in the grid, from left to right or X axis) and then to edit the nodes in the subtree? In this way I substitute the long connection with a very short connection to a subtree, and then a long connection to the child node that starts the real subtree.

The problem is when subtrees nodes, since they do nothing apart making positioning easier, are not counted for the priority determination

Well yes, if the system is not reactive, it is turning off. The challenge is to say "ok gladiabots cannot be much optimized, so can I make small and slick decision trees?".

Anyway I saw that you got grip, welcome in the 1400+ points club.

Yes but then the problem is not the elo system. Is that the "best tactic" for the map was found, That's it.

The draw is punishing in the elo (come on, getting upset for internet points -.- ) system because you get paired with someone that is way lower than you. If you would be paired only with guys over you, you will get points from draws.

It is like "wait, I got 1600, let me stop playing now because otherwise I may lose points". Make a screenshot and go on.

Ok, I play on the nvidia k1 and I do not have much problems. Surely if you have 50. I have 12.

Elo is not used in tournaments because it makes sense only for "infinite" competition. Draws in chess are daily routine especially at difficult tournaments.

Yes but on average you do not play only those maps. Plus I lose from time to time on mind game, there are some interesting tactics.

How high are you in the rankings? Or which elo do you have?

I win and lose on CoD. Not yet optimized. How many points do you have in game?

side note on your recent posts: if you draw too much it is boring. If you win too much it is boring, If you lose too much it is boring. :rolleyes:. In the case you win, it is a challenge for the others to at least try to draw! (since when they discover your same AI, it will be a draw, but first they have to find it!)

But the elo system is quite solid (and indeed it accounts for draw when one player should have won). I do not get the critic.

The elo system tries to predic the probability that you will lose/win/draw your next match. If you lose points with a draw, it means that for the elo it was expected that you would have won. Finding another solid system (that is not similar) is not trivial.

While it is true that the elo does not really fit when more than two players play a map. (imagine triforce)

Woah on an s6 is slow?.Anyway you may convert your ai in text so you can copy them back. It takes a bit of time, but it is doable.

Indeed, mine is more "just try". Not really an heuristic.

Cannot follow totally but kudos for engaging yourself in such games at 47!

Nice approach. Mine was a bit more complicated: given a maximum number of nodes, try to build (and test) the decision tree on some sample maps and save the top X (15 for example). But even before starting with 100 nodes I can wait years to let it finish.

I program things bed sloppily just enough that they run cobbled together with digital duct tape.

I know this too well. Well then if you are close I assume that you have a good approach (if it is valid). The way that I approached it, for every possible behavior or decision tree, gets too big and too quickly. Not even the space of possibilities of chess.

Old, almost too old. Over 25.

And you?

I do not know how far are you in computer science, how old are you and so on. Surely you leave some sentences that seems to point that you are quite "scientific" and able to program. But you should know how to approximate the "space of solutions" before starting such things. You will need likely years even with a supercomputer and an optimized algorithm. I already thought about this and it is quite an enterprise. You should also care if you keep every possible solution in ram, there would not be enough ram for it.

You may try with genetic algorithms, game theory, min max pruning (AI and heuristics), machine learning (for example: neural networks), etc.. and you will learn a lot of useful stuff (professionally, indeed gladiabots is awesome for this too) but as I said, it is quite an enterprise. It is more likely that a human mind will develop faster a better decision tree.

So, in any case, keep going, it will do well to you! Just, from time to time share! (I mean methods, not only results or sentences).

Great I would like to see how do they work. Since in league4 I still see a big K factor.

GFX47 I agree with the change but also with Johnbob. Maybe you can delay this for a while to not "stress" to much the playerbase. I myself was able to complete the draft of the new decision tree (that so far is not losing) only after 10 days.

At the end it is a little programming game, if the conditions always changes one becomes frustrated. Said that I like the idea (and it will need a lot of balance as well). I like to use assault bots and special bots in a "precious" way.

Yup the capturing mode of WOT is quite well known. Already quake had it.

In short: in the area for capturing there should be no enemies and the allies capturing the area should not be under shooting.

But I guess those modes will come. I would consider the game challenging enough for the moment. One has to optimize per mode, per map, per class. per enemy class.

So did you play with that AI most of the 4000+ games keeping your around 1600 points? Kudos!

I remember your stats. Like 4000+ games (what a patience!) and oscillating between 1550 and 1650. Today I was even able to beat you, I guess you are still testing.

woah never lost, quite a work. Share, let's see if we can refactor it together!

Cannot be that you already do what I think a lot of games should do since years. Too much awesomeness.

I did for the assault. I was not convinced by the new bot, now I started again but, yes, I see assault not so useful at the moment. It should be coded to have its sweet spot at medium range I think.

This is a request when the game gets a bit stable. It is frustrating to see a player that played in the past, got a lot of points, and then idles there in the leaderboard. This happens in a lot of games and it "turning off" especially when the game gets a bit old. It is like the leaderboard is full of inactive players and it feels the game is a ghost town.

Sometimes it happens than a game had certain conditions were players reached very high scores, that are not possible anymore (like when a game get crowded or depopulated), so they hold the top spots and that's it.

Would be nice if the players that did not play for the last X day (say 30), would be hidden. They may retain their score, but at least the leaderboard will show only the active players and so the actual user base strength in the game.

My current decision tree still in heavy test has: 33 conditions , 45 actions so 78 nodes.

See you at around 1500 points it is a good point to know that few good nodes are effective.

Thanks for sharing.

PS: seven wonders is my preferred map! One day!

But I mean useful AIs. If you have an Ai with almost 500 nodes but you do not use it actively in multiplayer, it does not count. Surely kudos for the patience! Could you report the nodes of your active Ai for multiplayers?

Anyway, woah, 500 nodes I wonder how detailed you got. Did it help?

As the question said. Actually to show how complex is an Ai/decision tree one should count the valid path, but let's count the nodes that are simpler. So how many actions + conditions (no subtrees) has your most effective (in multiplayer) AI?

disclaimer: it is not true that more = better, but surely it is an index of how much a player is engaged.

Not sure if I follow, could you elaborate?

Well why not.

Next in 5.3: autocombustion of bots, randomly assigned.

I'm really interested by those sentences of yours "probability analysis algorithm". I mean, I know what you mean in theory but i would really like to see what you use