Results 1 to 4 of 4

Thread: A moooaar complex algorithm

  1. #1
    Basic Member
    Join Date
    Apr 2012

    A moooaar complex algorithm

    Performance bar algorithm should be complex enough to take these things in consideration:

    - Total damage taken / done by a hero;
    - Total healing done;
    - Total damage mitigated;
    - Total damage done to towers;
    - Last hits;
    - Items bought;
    - TP's bought / used;
    - Key moments: Every game should record about 5 key moments, like a big teamfight, a tower dive or Roshan. During those moments a complete and detailed log of damage done / skills wasted / a good placed ulti / etc should be recorded.
    - The role in the specific game and how the above apply;
    - Etc.

    Some things are pretty straight forward, like tower diving to pursue a stunner and get killed or buying an Orchid and wasting it a lot on a Linken. Those should be adressed. Of course, there are odd exceptions, but that's why we need a good sample.

    Then we can start talking about the possibility of a not completely useless bar.

  2. #2
    Basic Member BZAD's Avatar
    Join Date
    Jul 2012
    This needs a picture of the cornflakes eating guy telling someone to do it right.
    ~Never forget~

  3. #3
    Basic Member DarkLite's Avatar
    Join Date
    Dec 2011
    Patch Anticipation Station
    Don't do silly performance bars. Give us actual numbers instead.
    Please help keep the forums clean by reporting rude or abusive posts!
    Got a problem? PM me or another moderator!
    Forum Rules

  4. #4
    Basic Member
    Join Date
    Dec 2011
    Instead of items bought, maybe item usage. If someone buys a dagon, how much total damage he adds with it? Dagon is not a bad item. Dagon can be used by all heroes (depending on how the match is developing). Rating performance lower because someone bought dagon on an agility/str hero shouldn't ever happen. It should be obviously more complicated than that: a clinkz with a dagon has a high chance of dying so he might not participate enough in engagements due to item choices. So in a match he might be ok with solo ganking and his team leaving him room to not need to tank up any AOE during engagements. In other matches, he will kill one support and contribute nothing in a push/counter-push.

    Items, skill builds, splitpush/safe farm/gank/push decisions are all too complicated... but some simplified statistics algorithm could take care of that. If you go dagon clinkz in 500 matches, and in 80% of them you make the best use of it[*], nobody can complain. So the performance bar will tell you that you're doing fine.

    Not sure how easy it would be to develop a metric to measure that though. I fear replay parsing would just get slower, since one of the first things that comes into my mind is using hero vision + distance from stuff (allies, enemies, enemy tower, creeps, neutrals, roshan, rune) to define "Affordability". So someone that is away from the team + enemies but near enemy tower + creeps can farm, do tower damage or tp back. You can check then the items and skills he have. A naked battlefury deals X dps currently: if he tp backs and just die in engagement, he didn't make use of anything he could "afford". So he gave up tower damage + farm affordability for enemy damage/damage mitigation/skill afordability. He gets a score of -5 in tower damage + farm affordability and -5 in the other one.

    Runs this as a time series, correlate it with mmr among heroes... there you have your algorithm. Every skill or item give the hero some adjustments to affordability. If the guy has vanguard, he can afford to splitpush if he tanks creeps + some tower shots to make a siege machine survive, for example. He can also afford some damage mitigation in engagements. Just a matter of measuring these.

Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts