Hive 3.0, com elo rework suggestions
Katzenfleisch
Join Date: 2014-03-21 Member: 194881Members, Squad Five Blue, Squad Five Silver, NS2 Community Developer
Join Date: 2014-03-21 Member: 194881Members, Squad Five Blue, Squad Five Silver, NS2 Community Developer
Hello,
After a lot of talks in the ns2 discord channel, I am creating a post here to see what can be done to address (or explain) the issues.
# Current system brief
Hive 3.0 introduced a specific elo set for commanders. When the round ends, the team elo avg is computed according to the time people spent in the chair, and on the field.
# Expected goals of the commander elo
It is known that commanding and fielding are not the same: Having a high elo player commanding was widely regarded as bringing an uneven round due to that player not playing on the field, where he is the best at. That high skill player could be kicked/ejected/blamed because of that reason alone. Hive 3.0 goal was to address that and allow high skill players to try themselves at the commander role without getting punished.
# The issue
The system is using field elo when a commander jumps out. As a Khammander it's not uncommon to jump out to defend, go gorge to echo and build rts, or just bank pres to buy a lifeform. It's part of the role and is done by skilled commanders. Sitting 100% in the chair is not the most effective way to command.
Also, a commander fielding is not the same as fielding as the job done is not the same. A marine will push, shoot, lane or ambush. As a commander you will build, protect your base against lone skulks or save gates while carefully watching the tech flow, and med requirements to jump back in.
Scenarii showing the issue:
* A 6500/300 elo marine commander and a 500/300 elo alien commander are playing. They both have 300 commander elo but during the round they will spend time on the field building stuff (arms labs, RTs, Harvesters, upgrades, etc). However, the system assumes the marine team has an advantage, which is impacting how elo is gained and loss thus defeating the elo system for every other players (elo will be less relevant).
* A 6500/300 elo marine commander is on the field 60% of the round. The marine team will be seen as hugely stacked, however, by doing so the commander is actually not capable of using any of the commanding tools (including the beacon, and base defending). Marine are expected to win with the system, which is not true at all. If the marine team loses, then that player will lose mostly field elo while it was a commanding issue, and the commanding elo should be adjusted.
* By jumping out as a high skill field player, you can only lose elo (since teams are marked as "stacked" if you win, you, or any member of team, won't gain much)
# My proposed fix
Fielding is part of the commanding job because by doing so you are doing a trade off. I recommand to apply the following:
1) Only use 1 elo for each player (commanding or fielding), use the commanding elo of the player with the most time in the chair/hive.
# Sandbox of ideas and arguments picked from the discord discussion (pro / against / food for thoughts)
- The system could balance it out once commanders have there elo setted up. A T7 on the field is still a T7, even if he commands and the hive should count that.
- Hive 3.0 could be set to the field elo, so that we keep how hive 2.0 works while making it improve over time by itself.
- Hive has to be able to deal with all kinds of messy situations, like different people hopping in and out of the chair all game, and things like that. And there's no way for it to tell "standard meta" from anything else. So a solution that handles all cases is to use your player skill when you're on the field and your commander skill when you're in the chair.
# Discord backlog
Long discussion starts here: https://discord.com/channels/84689542480674816/361232765460676610/772437120853475348
After a lot of talks in the ns2 discord channel, I am creating a post here to see what can be done to address (or explain) the issues.
# Current system brief
Hive 3.0 introduced a specific elo set for commanders. When the round ends, the team elo avg is computed according to the time people spent in the chair, and on the field.
# Expected goals of the commander elo
It is known that commanding and fielding are not the same: Having a high elo player commanding was widely regarded as bringing an uneven round due to that player not playing on the field, where he is the best at. That high skill player could be kicked/ejected/blamed because of that reason alone. Hive 3.0 goal was to address that and allow high skill players to try themselves at the commander role without getting punished.
# The issue
The system is using field elo when a commander jumps out. As a Khammander it's not uncommon to jump out to defend, go gorge to echo and build rts, or just bank pres to buy a lifeform. It's part of the role and is done by skilled commanders. Sitting 100% in the chair is not the most effective way to command.
Also, a commander fielding is not the same as fielding as the job done is not the same. A marine will push, shoot, lane or ambush. As a commander you will build, protect your base against lone skulks or save gates while carefully watching the tech flow, and med requirements to jump back in.
Scenarii showing the issue:
* A 6500/300 elo marine commander and a 500/300 elo alien commander are playing. They both have 300 commander elo but during the round they will spend time on the field building stuff (arms labs, RTs, Harvesters, upgrades, etc). However, the system assumes the marine team has an advantage, which is impacting how elo is gained and loss thus defeating the elo system for every other players (elo will be less relevant).
* A 6500/300 elo marine commander is on the field 60% of the round. The marine team will be seen as hugely stacked, however, by doing so the commander is actually not capable of using any of the commanding tools (including the beacon, and base defending). Marine are expected to win with the system, which is not true at all. If the marine team loses, then that player will lose mostly field elo while it was a commanding issue, and the commanding elo should be adjusted.
* By jumping out as a high skill field player, you can only lose elo (since teams are marked as "stacked" if you win, you, or any member of team, won't gain much)
# My proposed fix
Fielding is part of the commanding job because by doing so you are doing a trade off. I recommand to apply the following:
1) Only use 1 elo for each player (commanding or fielding), use the commanding elo of the player with the most time in the chair/hive.
# Sandbox of ideas and arguments picked from the discord discussion (pro / against / food for thoughts)
- The system could balance it out once commanders have there elo setted up. A T7 on the field is still a T7, even if he commands and the hive should count that.
- Hive 3.0 could be set to the field elo, so that we keep how hive 2.0 works while making it improve over time by itself.
- Hive has to be able to deal with all kinds of messy situations, like different people hopping in and out of the chair all game, and things like that. And there's no way for it to tell "standard meta" from anything else. So a solution that handles all cases is to use your player skill when you're on the field and your commander skill when you're in the chair.
# Discord backlog
Long discussion starts here: https://discord.com/channels/84689542480674816/361232765460676610/772437120853475348
Comments
Here is another example why the current comm skill is flawed.
Right now if you leave the commander slot to build structures like an armslab or to collect pres the game judges you by your field elo.
A commander who leaves his place is usually stronger than a 100% pure commander, but just by a slight amount.
Since the majority of player have more field hiveskill, the game will record the match as stacked in your favor by a huge margin,
in case for some sanjis even by +400 average hiveskill and expect you to win over 80% of your games.
Hiveskill 3.0 would work right, if all commander had their commander skill truely reflect their power like 2000-3000 hiveskill.
But right now you can abuse the system with a low hiveskill command skill by jumping out and collecting pres between researches.
Hiveskill 3.0 will calculate with your time doing nothing outside of the hive an expected +80% winrate and
it will lower your commander hiveskill over time until you reach this inflated winrate.
Some people argue its good to punish out of chair commanding with losing hiveskill, since high command hiveskill should only be possible with pure commanding,
but there are enough sanjis who like to win +80% of their games more than a high number.
Its a system which turns all out of chair commander to smurf in the future, even if they dont intend to be.
Possible solution to prevent it:
1. You could start commander skill at your average field skill instead of 0, but then some hightier player will never start to command.
(there are a lot of sanjis with <20h in commanding)
2. You could remove the option to jump out of the chair/hive, but it would remove a big aspect of the game.
(and what about relocating your chair/main hive dying?)
3. You could stop counting the default comm as field player no matter what he does, but pure commander will have an disadvantage over out of hive commander
(what about changing commander?)
The third solution would work close to hiveskill 2.0 with the difference that if the comm switch places with a field player the game will be stacked.
To solve this, you could force the commander role on the default comm, field player are just allowed to drop eggs/ips and maybe cysts.
The default comm has the task to do the commander duties or his team will probably lose.
There are still some issues with it like what happend when you eject your commander or how harsh you want to restrict field player from commanding,
But if the marine comm has a 6000 (field) point advantage and jumps out to help, that must be reflected in the system aswell...
On public you can expect a sanji to hold a 2 skulk baserush alone as a commander.
The problem I see is whether or not the commander actually jumps out. If you have a sanji comm, you MUST expect him to jump out (just as you were expecting him not to comm with H2.0).