|
This will be the center of discussions for solutions to the following problem:
SQ is being eliminated by HotS and it has the weakness of not being able to be accessed from a replay, preventing a working Sc2 gears plug-in and the mass tournament replay analysis that would be necessary to make it a super accurate and up to date metric.
The goals are as follows:
Create a metric that measures macro using data we can get from a replay file (SQ).
Conduct massive ladder analysis and pro tournament analysis to bring SQ up to date.
Create a Sc2 gears plug-in that will give you your SQ for every game.
Issues:
How do we get replays that we can guarantee are from a certain league? Can people who upload replays to sites like Sc2rep.com say that they're in diamond league when they're in silver?
I've contacted WhatTheFat as well as a few other people who I believe will be interested in this project. Please leave comments with your ideas etc
GLHF,
JaK
  
|
My idea:
I think we can still use average income and average unspent resources because the data for the amount of minerals and gas is in the replay (right?). Someone tried to do this while making a custom map for TheStaircase and said that for some reason his numbers didn't match up with the exact numbers on the score screen. For our purposes, this doesn't matter, as long as its consistent we can re-analyze everything with these numbers and adjust the SQ Scale accordingly to accurately reflect the average SQ levels of each league.
I also think that supply blocks and supply efficiency are important, as well as energy management (inject, chrono, and mule efficiency). Can we make 3 scores, one for spending, one for energy management, and one for supply then make them into a combined score for a total overall macro score? Are there other critical aspects of macro that I am missing, maybe worker production for T and P?
|
The concept of energy management is a very tricky one. Chrono is barely used past midgame, Queens sometimes don't Inject so they can save up energy for Transfuse, and Terran saves Scans frequently. However, it's clearly relevant, mostly for Zerg. Maybe including Chrono could be a good plan, but Terran should be disregarded for energy management because of saving Scans.
Supply is something that was a big flaw in SQ. Being maxed for long enough decreases your SQ drastically. Therefore, we need to make that portion count. Simply disregarding all information on spending gathered when maxed isn't the solution at all. Instead we need to think of some statistic that would compensate for being at 200/200. It could be anything, like Army Unit Spending, then making an equation which uses it so that when Army Unit Spending is lower, then the spending score is less affected. There are a few things wrong with Army Unit Spending, such as the exploit of queuing up units, but it's just an example.
|
finding a way to solve the exploit of queued up units is definitely on the list of things to do with this project.
|
afaik replayFu indicates the league through Bnet. It's not a very used website tho
|
On November 16 2012 07:56 loginn wrote:
afaik replayFu indicates the league through Bnet. It's not a very used website tho
nice one, that should work fine, we only need a sample of players.
|
Technically, everything is available in replays, the question is how much work it is to extract, the simple one being just read a number telling you something (average income etc), the complicated one being to simulate the game from the replay and get every single bit of information. I think we're all forced to make compromises since no one is going to literally write a simulator for the replays, but then again, question is how much information is needed and how useful it is with too much information. We all agree that energy consumption is useful, but at the same time, if you include too much data in SQ, the rating will become very complex and hard to understand. "Crap I got a low SQ, but I'm not sure why" etc.
|
On November 16 2012 17:11 Tobberoth wrote:
Technically, everything is available in replays, the question is how much work it is to extract, the simple one being just read a number telling you something (average income etc), the complicated one being to simulate the game from the replay and get every single bit of information. I think we're all forced to make compromises since no one is going to literally write a simulator for the replays
You don't need to simulate a game in order to acquire information. SC2gears gets amazing amounts of information from replays without doing that.
but then again, question is how much information is needed and how useful it is with too much information. We all agree that energy consumption is useful, but at the same time, if you include too much data in SQ, the rating will become very complex and hard to understand. "Crap I got a low SQ, but I'm not sure why" etc.
No, I think he means breaking it down and showing all of the major sections of the macro rating (it's not going to be called SQ because it's not just spending) then show the total macro rating which results from them. That way you can simplify it if you need to.
|
On November 16 2012 21:23 Antylamon wrote:Show nested quote +On November 16 2012 17:11 Tobberoth wrote:
Technically, everything is available in replays, the question is how much work it is to extract, the simple one being just read a number telling you something (average income etc), the complicated one being to simulate the game from the replay and get every single bit of information. I think we're all forced to make compromises since no one is going to literally write a simulator for the replays You don't need to simulate a game in order to acquire information. SC2gears gets amazing amounts of information from replays without doing that. Show nested quote +but then again, question is how much information is needed and how useful it is with too much information. We all agree that energy consumption is useful, but at the same time, if you include too much data in SQ, the rating will become very complex and hard to understand. "Crap I got a low SQ, but I'm not sure why" etc. No, I think he means breaking it down and showing all of the major sections of the macro rating (it's not going to be called SQ because it's not just spending) then show the total macro rating which results from them. That way you can simplify it if you need to.
Yes, it gets a ton of information because the information is there in the replay file. The problem, apparently, was that it isn't anymore in hots. If it doesn't say anywhere in the replay file how much minerals was gained, for example, you would literally have to simulate/calculate using the amount of workers built, the paths they took to which crystals, how they were moved during engagements etc, or the information you extract is bogus and doesn't properly represented what actually happened during the game.
|
I don't think the replay files are changing, just the score screen (where SQ currently comes from).
|
JaKaTaK United States. November 16 2012 05:19.
... Can we make 3 scores, one for spending, one for energy management, and one for supply then make them into a combined score for a total overall macro score? ...
Don't over complicate stuff when you don't have to. 3 Different scores* give you more information than one arbitrarily weighted combined score*.
I don't have a clue about the available information of replays and the map editor, but I guess that it's rather easy to count the time during supply blocks (while the player is not maxed) in a funmap. I fear the different Supply costs for units are kinda a problem. Maybe someone got 95/100 Supply so that he can't build his Thor/Carrier/Ultralisk. Maybe he wants to build 5 of those units at the same time but he doesn't start (some of) them because of the Supply block (or queues them up (P and T))... Anyone has an idea how to solve/avoid that problem? Considering someone to be supply blocked while he still could build units based on his ability to build units (production capacity) doesn't sound like a good idea, cause you don't want to build constantly units from every production building in many cases (4+ Colossi, Medivacs/Vikings, adding Hydras in ZvZ, ...)
Maybe calculate the "wasted" Energy, while each Nexus/OC got max Energy?(Cumulated_time_of_max_Energy_for_all_Headquaters*Energy_regeneration_per_second)
Which information are exactly available (in replays)? What does HotS measure in it's scores screen?
+ Show Spoiler +*[I am not sure about the English Vocabulary but shouldn't it be "key figure"?]
|
On November 18 2012 09:30 Uncreative_Troll wrote:Show nested quote +JaKaTaK United States. November 16 2012 05:19.
... Can we make 3 scores, one for spending, one for energy management, and one for supply then make them into a combined score for a total overall macro score? ... Don't over complicate stuff when you don't have to. 3 Different scores* give you more information than one arbitrarily weighted combined score*. I don't have a clue about the available information of replays and the map editor, but I guess that it's rather easy to count the time during supply blocks (while the player is not maxed) in a funmap. I fear the different Supply costs for units are kinda a problem. Maybe someone got 95/100 Supply so that he can't build his Thor/Carrier/Ultralisk. Maybe he wants to build 5 of those units at the same time but he doesn't start (some of) them because of the Supply block (or queues them up (P and T))... Anyone has an idea how to solve/avoid that problem? Considering someone to be supply blocked while he still could build units based on his ability to build units (production capacity) doesn't sound like a good idea, cause you don't want to build constantly units from every production building in many cases (4+ Colossi, Medivacs/Vikings, adding Hydras in ZvZ, ...) Maybe calculate the "wasted" Energy, while each Nexus/OC got max Energy?(Cumulated_time_of_max_Energy_for_all_Headquaters*Energy_regeneration_per_second) Which information are exactly available (in replays)? What does HotS measure in it's scores screen? + Show Spoiler +*[I am not sure about the English Vocabulary but shouldn't it be "key figure"?]
The three scores will be shown separately, then combined. If I am not mistaken, the combined score is supposed to help you compare your macro to the macro of other players, while the separate scores are meant to help you find out what you did wrong.
I don't see the problem in higher supply units, even with the explanation. Higher supply units generally cost much more than 5 marines, mostly because of gas.
The wasted energy idea is an interesting concept which would solve many problems with Toss and Terran concerning energy management, while at the same time promote Toss players to use Chrono more often.
All I know about replays is that they store every single command carried out by all players, from clicks to keystrokes, excluding the ones having to do with options and menus. I think they also store data about minerals, gas, and supply as well, considering what SC2gears can do. The score screen in HotS is essentially the same as the one in WoL except it is missing a couple statistics and has a couple new statistics.
|
Antylamon United States. November 18 2012 15:08. Posts 1095
...
I don't see the problem in higher supply units, even with the explanation. Higher supply units generally cost much more than 5 marines, mostly because of gas.
...
+ Show Spoiler +Uncreative_Troll November 18 2012 09:30. Posts 78
I don't have a clue about the available information of replays and the map editor, but I guess that it's rather easy to count the time during supply blocks (while the player is not maxed) in a funmap. I fear the different Supply costs for units are kinda a problem. Maybe someone got 95/100 Supply so that he can't build his Thor/Carrier/Ultralisk. Maybe he wants to build 5 of those units at the same time but he doesn't start (some of) them because of the Supply block (or queues them up (P and T))... Anyone has an idea how to solve/avoid that problem? Considering someone to be supply blocked while he still could build units based on his ability to build units (production capacity) doesn't sound like a good idea, cause you don't want to build constantly units from every production building in many cases (4+ Colossi, Medivacs/Vikings, adding Hydras in ZvZ, ...)
I tried to explain that someone can be "supply blocked" while still having free supply to build units. You won't always reach N of N supply because not every unit costs 1 supply.
The unit with the lowest supply (min(S)) which is currently build doesn't have to be 1. In this case it's possible that you won't be able to reach N of N supply but N - X supply with X<min(S). (Because of this you can't start the timer only while the player has N of N supply. Starting earlier might cause horrible results.)
A Zerg might be considered supply blocked, when he couldn't spent all his Larva (but wanted to), but you sometimes might want to save up Larva/wait for Tech/Scout so you can't use Larva to measure supply blocks...
|
The definition of a hard supply block as I see it is:
A point in time where you want to build a unit, but cannot, solely due to the fact that you do not have enough supply for that unit, and you will not have that supply within the time it takes to build 1 overlord, pylon or supply depot.
The tricky part it "where you want to build a unit". There will likely be times where you are intentionally supply blocked, like when a protoss is 26/26 during a 4 gate. However, I don't think it would be a bad thing to know the number of times I am hard supply blocked in a game without having to watch the replay and count them, or try to remember the exact number. I can say, "okay, in this build I will be supply blocked at this time, and its all part of the plan, so I should have 1 HSB in the analysis at the end of the game." If the number is 2 or more, I know that I made a mistake somewhere, and it should be easy to show the exact time when this happens (the apm tracker already does this).
|
Some notes on this. Average Unspent Resources and Collection Rate are not provided in the replay, they are in a .s2gs file that the game generates from the reply by replaying all the actions. http://ggtracker.com though has the ability to get these files (likely but opening them in a replay viewer or something of that sort, they have not gone into details as to how they are doing it. They suspect that things will work fine with HotS but are unlikely to release the feature for HotS replays until HotS itself is released.
Getting these numbers in a custom map is very possible by keeping track yourself, though it may drift due to sample rate being a little different. Unfortunately in order to get any data just purely from a replay for SQ, supply blocks, energy, etc you have to simulate the game to varying degrees.
|
Actually, Dave from ggtracker said it would take him 4 days work to create HotS compatibility. And he can get the average collection rate from the graph on the score screen, as it is the same as WoL. I convinced him that HotS compatibility is a very important thing, so we should be seeing some updates from ggtracker soon. 
|
Ah cool, I had talked to him early last week which is what my post is based off of. Great to hear.
|
Hi! So, indeed the rumors are true and I'll be add HotS support to GGTracker, I hope within 3 weeks. And yes, resource collection rate can be computed from the s2gs / score screen, and it is _exactly_ the same as WoL.
|
Well it can't be computed from the score screen like in WoL because it isn't given as an exact but, just as a random point at the end of the graph. I am glad to hear it is still in the HotS s2gs file.
Edit: because I was having a bad day
|
@Wraithan
Woah, harsh dude. I'm sure he's got a shit ton of things on his plate aside from adding HotS support to GGTracker. Dude does an awesome job, have some respect.
|
|
|
|
|
|
EPT
