Improving bot mid behaviour

[Django]

tldr: mid bot should not roam so much, rather compete vs the enemy in mid lane

List of heroes that start roaming at level 2, being an utter fail in the mid lane (you should never do that, basically stay in lane and only gank if you have a good rune + a pushed enemy lane; otherwise you give the enemy mid freefarm + xp, very bad)

- Tiny, Lina, Venge, Zeus, Admiral, Windrunner

They leave mid at some point and form a trilane. They maybe come back but usually underleveled and underfarmed if their "gank" failed

List of heroes that stay in mid and try to lasthit (the only ones playing a decent mid)

- Sniper and Razor (maybe because tagged as carries?)

The rest I have never seen mid or don't remember.


What the bots actually should do:

1) Push the lane with spells at x:45-x:50, -then- go check for runes

2) Stop ganking the sidelanes so early (most of the time its not even a gank, it's just forming a trilane); exception might be a good early rune like haste

3) get XP / lasthits / denies until ~lvl 6

4) then start ganking (through woods, not through river) the sidelanes, or farm some more

5) make the sidelane bots gank the mid from time to time (with heroes like venge, cm, es or tide)

[Yorien]

I believe that currently the bots who tend to roam instead of staying midlane farming are the ones that have any sort of stun/cc ability, so they can perform a more successful gank. Bots without stun mechanics will stay in lane farming till their lane phase is over.

[squinte]

Test this out on Normal and Unfair.
Bots who are in mid that tend to roam allow you to free farm for way too long. They won't return for several minutes allowing you to become much stronger than them.
On Normal this is true. on Unfair, they are able to stay on par with your leveling because of the exp boost.

However, it's not uncommon for the mid laner to roam to gank a side lane. It's sort of a problem in pub matches where nobody wants to gank another lane until like 15-20 minutes in, more so in the less skilled regions of pub matches. That doesn't mean mid shouldn't be trying to gank. If they've got the tools, and farming mid allows them to get those tools quicker, the mid should be trying to initiate a gank on another lane because they will be the strongest heroes on the field.

I agree that they do try to gank far too early into a match though. It should be at least 2-4 minutes in before mid starts to try to gank another lane depending on runes and how out leveled you have become versus the enemies in a specific lane. Even so, out numbering the enemy 3-to-2 should be enough to have a successful gank.

[Django]

1. yes i do play on unfair too. still, the xp boost does not matter, because the AI often stays very long at the lane --> trilane (xp split thrpugh 3)--> very poor xp gain, im still higher in mid lane than the bot

2. I'm not saying ganking the sidelanes is wrong. I'm talking about a mid AI that is missing 80 % of the time during the early laning phase. That's not ganking anymore, that's roaming. You should never do that as mid unless you play dual mid.

3. Actually the problem of pub games is that the sidelanes refuse to help out the mid player. Mid's job is not always ganking. It depends on the hero played.
In high level games the sidelanes gank the mid, at early levels (when the xp difference is not that big yet), that's how it's done.

Take Tiny for example: he needs xp advantage over other heroes on the map too knock them out with his combo succesfully. thats why he stays in mid until level 8 and then starts roaming around. (of course tiny can gank earlier, but this usually requires a rune - and it is still better to take out the enemy mid with that rune usually)

New players might copy that AI style but try abandon the mid lane in a real game and fool around on the sidelanes early on, it's a guaranteed loss if you have a decent opponent mid.

tldr: mid should stay at it's lane early on. make sidelane bots gank the mid instead. that would be really challenging. as human player I had to be constantly aware of the map and look out for a venge or es stun.

[squinte]

Right, the problem I've noticed is that when more than 1 bot shows up in mid, it's usually like 12+ minutes into a match and it's never to gank. Usually they're there to make a team based push. They're looking for a team fight more than anything else on that point. Usually by that same token, the opposing teams outer lanes are still stuck in farming mode and won't come mid to defend because they either got out farmed or ganked too many times early game and are stuck in the farming loop.

As for high level games and ganking mid lane, I don't necessarily agree. It really is dependent on the hero in that specific lane. What I mean is, sure they could gank mid, but why bother ganking Shadow Shaman in mid lane and let Anti-Mage free farm in bot lane? Clearly the priority should be on AM, and I think the bots need to be able to recognize which hero on the opposing team needs the most farm and try to make their lives a living hell, especially on unfair. Whether that means having mid go for a rune gank or have the bottom lane leave to both gank mid.

[ChrisC]

Having the bots focus on ganking particular heroes rather than whoever happens to be in a lane they decide it's opportune to gank is a good suggestion.

That said, it's definitely harder to gank mid, just because of how the map's laid out around there.

[Django]

Does that mean we will have still have roaming bots mid?

[Terin7]

Having side line bots gank mid should probably be a pretty low priority function. Easy to focus on improving the mid bot's ganking behaviors. The best quick fix for now would be to have the bot not really gank until 5-6. Then, focus on having the bot try to get rune control (via pushing the lane/harrassing the lane opponent back, not just walking to the rune once it's spawned), then determining whether it would be possible to gank with said rune. That could be based on the type of rune (haste usually no brainer, regen if it can make the difference between having enough nuking power) and the state of the lane (using an invis rune only if the enemies are far enough from a tower, taking quick path if enemies have no escape mechanism, or longer route to flank if they do).

As a semi-unrelated piece of feedback that sort of ties into sidelanes ganking mid, perhaps bots should have a "State of the lane" that could help them have more dynamic actions throughout the laning phase? Making use of gold and XP per minute on a lane-by-lane basis would be a sort of simple way of determining which team is "winning" which lane. For instance, say there's a bot vs. human team. The human mid got first blood on the bot mid, and has been denying a decent amount, so the bot mid is underleveled and the human mid has a gold advantage. However, the bot top lane got the second kill in the game, and has a good level advantage, while the bottom bot lane is keeping neck and neck. This relates back to having bots roam mid to gank, then. So you could determine since the top lane is winning their lane, they can afford to send a hero to gank mid to try and make up for the deficit the mid bot has created.

This could also allow for some other features, such as having bots switch lanes if they are getting beaten constantly, or for quick pushes if the lane is susceptible to it (for instance, if an enemy lane has a Lich and Anti-Mage that is doing well, pushes are something they are vulnerable to, due to Sacrifice, that would make up for lost XP and gold). A mid hero would be able to make smarter decisions as well about ganking, to see is he should be incredibly aggressive against one lane, generally ganking all the lanes, or staying in his lane to try and win it, if that hero farming would be the biggest threat. This code would become mostly irrelevant past the first 10 minutes and laning stage, as heroes will be constantly roaming across the map. However, it could become more bare bones and be used to help determine pushing? Having weak, or an absence of, heroes, and how far creeps are pushes could determine how much the lane is "winning", could let bots think how much much will be gained by taking a tower (thus increasing how much the lane is won by), or prevent too much loss through defending a weak lane (compensating in this case, like a bot roaming to gank an enemy who is doing well).

Parts of this suggestion begin to fall into the category of having predictions, IE what would happen if we let that hero free farm vs. if he was pressured or ganked, which becomes a lot more tricky, but I think it could give bots a real edge in the laning stages, outside of trying to have them just work with the heroes they have in a lane. Dota is a team game, and thinning the barriers between lanes can significantly improve the bots in some situations. But for now! Make the mid bot not gank until 5-6, then possibly soon have them work on rune control, like the OP suggests. I personally would like to see a larger bot hero pool before something like my state of the lane suggestion, but whatever works best for you. Even if it's just a simple implementation of it, it could help the mid bot determine where he should gank and how hard.

[ChrisC]

Good suggestion on the "state of the lane" stuff, lots to think about there. Probably more of a long-term feature though.

And yeah I'll be working on tuning the mid roaming stuff. Part of it is just because the hero pool is so small right now, sometimes you get stuck with someone mid who really does want to roam and gank a lot rather than stay in lane (like Vengeful or something).

After getting them using the courier (pretty much the last standalone "feature" on my list), I think they'll probably in a good enough state overall that I'm going devote some time to try to get a new bot hero released each week.

[BLABLAFU]

Maybe opening a thread each week about the bot you want to implement next to gather some important things from the community on what to take care of to make the hero work properly may be a good idea.

Brings some involvement from the forumers and lessens the problem that you may miss something important.

[Django]

After getting them using the courier (pretty much the last standalone "feature" on my list), I think they'll probably in a good enough state overall that I'm going devote some time to try to get a new bot hero released each week.

That's some big news. Awesome.

[ChrisC]

Well, no promises, but I think that's probably pretty realistic.

[Terin7]

That's pretty awesome news that it can be done that quickly (for the most part). That would be a great way to get people involved with testing bots more, to see how the new bot works out, just like with hero releases.

[Vimsey]

Yes Ive noticed Tiny always goes on a magical mystery tour way too early and if he isnt successful you can get a massive lead. Its actually easier to play mid in a co-op game than solo on easier.

Ive noticed new players get pretty frustrated by being ganked by him though and often ragequit even if you are getting massively farmed and are about to make tiny pay for all his wasted time. But thats another issue. The plus side of this is that they soon learn to call MIA.

A regular thread we can post thoughts in would be a good idea I often dont feel its worth starting one of my own for a small subject.

Dont worry we wont hold you to it....much