Print

Please login or register.
1 Hour 1 Day 1 Week 1 Month Forever
Login with username, password and session length

[Blueswan]

According to Andrews log search I beat the bots less frequently than I thought. I almost always play either Conqueror Bot, Lord Bottington, Defender Bot or Banker Bot (Villager Bot and Warlord Bot tend to be lower ranked so I avoid them).

My winning percentage against those are:

Banker Bot - 75,4%
Lord Bottington - 74,2%
Conqueror Bot - 71,4%
Defender Bot - 65,3%

This is consistent with my feeling that Defender Bot is the one causing me the most trouble.

However, when playing with colonies - which happens far too rarely, my winning percentage against those four bots is upped to 85,2%.

[Ratsia]

My general feeling is that the bots are weaker in colony games, where they tend to go straight for Provinces whenever possible, as if it was a non-colony game.

More generally, the bots are weaker when the kindgom suits a strong engine or other complicated strategy. For simple kingdoms they indeed play reasonably well, both because it is inherently easier, but also because the bots probably have been handcrafted to play certain strategies with some skill (BM, gardens rush etc).

Whether playing against the bots makes the rating artificially too high or low depends on the set of kingdoms owned not just by you but by people playing against the bots in general. If you only have Base but most others own all sets then playing against the bots will make your rating artificially low. This is because the guys playing complex kingdoms will thrash the bots' ratings whereas you have to play them on kingdoms they actually know how to play. If you have all sets but most people play against the bots using Base only, then your rating will be artificially high; you beat them easily, but their rating reflects primarily their skill on simple kingdoms that you do not see that often.

The effect "the more complex kingdoms you tend to play (i.e., the more complex sets you own), the more your rating will be inflated if you play just against the bots" should hold in any case, but whether the spectrum goes to both sides of zero depends on the overall distribution of sets owned by people playing against the bots.

[Blueswan]

My general feeling is that the bots are weaker in colony games, where they tend to go straight for Provinces whenever possible, as if it was a non-colony game.

More generally, the bots are weaker when the kindgom suits a strong engine or other complicated strategy. For simple kingdoms they indeed play reasonably well, both because it is inherently easier, but also because the bots probably have been handcrafted to play certain strategies with some skill (BM, gardens rush etc).

Whether playing against the bots makes the rating artificially too high or low depends on the set of kingdoms owned not just by you but by people playing against the bots in general. If you only have Base but most others own all sets then playing against the bots will make your rating artificially low. This is because the guys playing complex kingdoms will thrash the bots' ratings whereas you have to play them on kingdoms they actually know how to play. If you have all sets but most people play against the bots using Base only, then your rating will be artificially high; you beat them easily, but their rating reflects primarily their skill on simple kingdoms that you do not see that often.

The effect "the more complex kingdoms you tend to play (i.e., the more complex sets you own), the more your rating will be inflated if you play just against the bots" should hold in any case, but whether the spectrum goes to both sides of zero depends on the overall distribution of sets owned by people playing against the bots.

According to that theory my rating should indeed be inflated as I have all the sets.

This needs to be put to a test. I think either on thursday or friday I will have to set aside a few hours to play against human players on my laptop. People can make bets however they like.   

I would probably bet on my rating going down, not necessarily due to the issues discussed here, but due to the fact that I tend to play too fast (=thoughtlessly) when playing humans (I don't like people waiting on me to make a play).

[Watno]

Numbers can be metaphoric?

Also, I find tat when people say 99% this is virtually never accurate.

Or to phrase it another way: 99% of the time people say 99%, it is significantly inaccurate.

But if the number is metaphoric,m the argument doesn't work anymore. If you win 90% instead of 99% of the time, the drop from a loss must only be 10 times as high as the gain from a win instead of 100 times.

[Obi Wan Bonogi]

I'm a noob, can someone explain the difference between the two different ratings? 

[ragingduckd]

I'm a noob, can someone explain the difference between the two different ratings?

6k is 1000 points higher than 5k. 

[ftl]

I'm a noob, can someone explain the difference between the two different ratings?

Between casual and pro, you mean? Pro is full-random and you don't know the kingdom until the game starts, casual you get to see and/or pick which cards you're playing with before the game starts

[Titandrake]

Also, the casual rating has a lot more uncertainty (you gain/lose a lot more points each game) I have somewhere around a 4.4k casual rating. I beat a 3.6k player, and win 100 points. A comparable rating difference in pro gives ~30 points for a win.

One particularly memorable game was where I happened to have a hot streak of casual games, and got up to a 7k rating. I lost a game to a 3k casual player, and lost 1000 points.

[Ratsia]

Also, the casual rating has a lot more uncertainty (you gain/lose a lot more points each game) I have somewhere around a 4.4k casual rating. I beat a 3.6k player, and win 100 points. A comparable rating difference in pro gives ~30 points for a win.

Have you played a lot more games for Pro? If yes, your rating estimate should be more accurate for that, which corresponds to smaller change in the rating. Hence, your observation could well hold even if the rating systems worked exactly the same.

That said, it would sound quite reasonable for the Casual rating to really be more dynamic.

[Vermillion]

According to Andrews log search I beat the bots less frequently than I thought. I almost always play either Conqueror Bot, Lord Bottington, Defender Bot or Banker Bot (Villager Bot and Warlord Bot tend to be lower ranked so I avoid them).

My winning percentage against those are:

Banker Bot - 75,4%
Lord Bottington - 74,2%
Conqueror Bot - 71,4%
Defender Bot - 65,3%

This is consistent with my feeling that Defender Bot is the one causing me the most trouble.

However, when playing with colonies - which happens far too rarely, my winning percentage against those four bots is upped to 85,2%.

Can you stand by until ragingduckd agrees to the terms as stated? He sort of did but I want to be sure he's down to clown.

Thanks for being a good sport in this.

[Vermillion]

My general feeling is that the bots are weaker in colony games, where they tend to go straight for Provinces whenever possible, as if it was a non-colony game.

More generally, the bots are weaker when the kindgom suits a strong engine or other complicated strategy. For simple kingdoms they indeed play reasonably well, both because it is inherently easier, but also because the bots probably have been handcrafted to play certain strategies with some skill (BM, gardens rush etc).

Whether playing against the bots makes the rating artificially too high or low depends on the set of kingdoms owned not just by you but by people playing against the bots in general. If you only have Base but most others own all sets then playing against the bots will make your rating artificially low. This is because the guys playing complex kingdoms will thrash the bots' ratings whereas you have to play them on kingdoms they actually know how to play. If you have all sets but most people play against the bots using Base only, then your rating will be artificially high; you beat them easily, but their rating reflects primarily their skill on simple kingdoms that you do not see that often.

The effect "the more complex kingdoms you tend to play (i.e., the more complex sets you own), the more your rating will be inflated if you play just against the bots" should hold in any case, but whether the spectrum goes to both sides of zero depends on the overall distribution of sets owned by people playing against the bots.

However, there bots play with pre-written scripts. I've played enough so that I can anticipate what script they are going to use when I see the Kingdom. And it's easier to learn these scripts - and the counters to them - if you limit yourself to playing the bots using the base set. Especially since the bot scripts have a huge weakness: they do not adapt/react (as best I can tell). But being able to do this is only possible if you are a pretty good player.

Most of the people playing the bots with base are new/less experienced players who aren't doing a couple games a day. Like my mom. Bots are going to chew them up.

Thus my theory is that bots are a conduit of points between casual newbs and decent players whose scores are overinflated. It's almost like there's a whole other ladder/point system of those who play just bots.

[qmech]

I'm around 6300 and typically play people around 5000, and 100+ point pro swings are about what I expect because I'm not playing very often.  So it could easily just be your higher uncertainty if you don't play many casual games.

[ragingduckd]

But here's the wager I offer:
1) He must play humans for 30 games straight.
2) He must how the game with a game title that is nondescript and must play whomever joins.
3) He may not pre-arrange any games with anyone.

If his score is LOWER than it is now, I win. If it is higher, you win. If it is tied, there is no winner.

Stakes: A physical set of guilds mailed to the winners home and a digital copy as soon as reasonably available.

I'm in, but I want to add a (4). We have to adjust somehow for Goko rating drift. I can detect it, since I'm scraping the leaderboard and I can match ratings changes with games and exclude the spontaneous ratings changes.

Of course, if bluswan plays 30 games in one night then it won't be a factor.

[Blueswan]

Ehhh....we're gonna have to postpone this. I just tried a couple of games on my laptop (against bots, as usual). Lots of lagging and the second game even froze up completely. Apparently, games work fine on my laptop, but NOT at home. So I guess the issue really must be with my home network after all, since I can play from my laptop just fine from my workplace. But obviously I'm not gonna play a load of games at work. 

Need to figure out what this network issue is about.

[yed]

Try HTTPS: https://play.goko.com/Dominion/gameClient.html

[Blueswan]

Try HTTPS: https://play.goko.com/Dominion/gameClient.html

Thanks, but it doesn't change the lagging, unfortunately.

[Vermillion]

But here's the wager I offer:
1) He must play humans for 30 games straight.
2) He must how the game with a game title that is nondescript and must play whomever joins.
3) He may not pre-arrange any games with anyone.

If his score is LOWER than it is now, I win. If it is higher, you win. If it is tied, there is no winner.

Stakes: A physical set of guilds mailed to the winners home and a digital copy as soon as reasonably available.

I'm in, but I want to add a (4). We have to adjust somehow for Goko rating drift. I can detect it, since I'm scraping the leaderboard and I can match ratings changes with games and exclude the spontaneous ratings changes.

Of course, if bluswan plays 30 games in one night then it won't be a factor.

You started this by asserting that he was 1000-1500 point below where he should be, and all I am asking for is 1 point. So at a minimum, I have spotted you 999 points. So I think the drift has been accounted for.

Is it possible it would work for him if he had a different browser?

[ragingduckd]

You started this by asserting that he was 1000-1500 point below where he should be, and all I am asking for is 1 point.

Who hey what now? 

[Vermillion]

You started this by asserting that he was 1000-1500 point below where he should be, and all I am asking for is 1 point.

Who hey what now? 

I misread your earlier post, but you were pretty confident. As for drift..... this was all about points, all things accounted for. If you want to increase the 'draw' range, what do you suggest?

[ragingduckd]

You started this by asserting that he was 1000-1500 point below where he should be, and all I am asking for is 1 point.

Who hey what now? 

I misread your earlier post, but you were pretty confident. As for drift..... this was all about points, all things accounted for. If you want to increase the 'draw' range, what do you suggest?

Here's the problem. As you can see, drift is a pretty severe problem, much more severe than I would have guessed. This is a way better bet for me (and way worse for you) if blueswan plays all 30 games in one 24-hour period. Honestly, I'm no longer convinced that there's any really fair way to construct such a bet on Goko, nor that "underrated" is even a well-defined term.

That said, I'd be happy to make the bet if we judge by rating relative to someone else who is also drifting, say the average of the bots. So if blueswan is 5000 and the bot average is 3500 when he starts his 30 games, then we decide based on whether he's more or less than 1500 over the bot average when he finishes, regardless of how long it takes him.

We have to change the stakes too, since I already bought Guilds online.

[Vermillion]

You started this by asserting that he was 1000-1500 point below where he should be, and all I am asking for is 1 point.

Who hey what now? 

I misread your earlier post, but you were pretty confident. As for drift..... this was all about points, all things accounted for. If you want to increase the 'draw' range, what do you suggest?

Here's the problem. As you can see, drift is a pretty severe problem, much more severe than I would have guessed. This is a way better bet for me (and way worse for you) if blueswan plays all 30 games in one 24-hour period. Honestly, I'm no longer convinced that there's any really fair way to construct such a bet on Goko, nor that "underrated" is even a well-defined term.

That said, I'd be happy to make the bet if we judge by rating relative to someone else who is also drifting, say the average of the bots. So if blueswan is 5000 and the bot average is 3500 when he starts his 30 games, then we decide based on whether he's more or less than 1500 over the bot average when he finishes, regardless of how long it takes him.

We have to change the stakes too, since I already bought Guilds online.

We aren't trying to make the wager fair, per se, it only has to be acceptable to you such that you think the pot odds are right for you. And the discussion WAS about points as it started. And if you think that accounting for drift favors me, then it's still a good bet for you. I still expect him to drop like a rock.

BTW - the wager was not 30 games in 24 hours. But 30 games. Possibly within a week or less?

And yeah, I bought Guilds, too. $4, not bad. I jumped on it because I Figured it was a pricing mistake that Goko would fix.

[ragingduckd]

We aren't trying to make the wager fair, per se, it only has to be acceptable to you such that you think the pot odds are right for you. And the discussion WAS about points as it started. And if you think that accounting for drift favors me, then it's still a good bet for you. I still expect him to drop like a rock.

BTW - the wager was not 30 games in 24 hours. But 30 games. Possibly within a week or less?

I wish I could respond more constructively, but I'm honestly not sure which of my posts you're misreading now. Maybe all of them.

We aren't communicating effectively, we can't agree on terms, and blueswan has technical difficulties... how about we just call this off?

[Vermillion]

We aren't trying to make the wager fair, per se, it only has to be acceptable to you such that you think the pot odds are right for you. And the discussion WAS about points as it started. And if you think that accounting for drift favors me, then it's still a good bet for you. I still expect him to drop like a rock.

BTW - the wager was not 30 games in 24 hours. But 30 games. Possibly within a week or less?

I wish I could respond more constructively, but I'm honestly not sure which of my posts you're misreading now. Maybe all of them.

We aren't communicating effectively, we can't agree on terms, and blueswan has technical difficulties... how about we just call this off?

The one where you referenced the ratings of of his computer opponents is what I misread/quoted.

I understand what you're saying re: point drift but I don't think it would affect him as much since he'd be playing every day or so. And it seems incosistent/difficult to calculate.

But if you want to withdraw, that's fine. Maybe when he can play again he will let us know.

[ragingduckd]

The one where you referenced the ratings of of his computer opponents is what I misread/quoted.

I understand what you're saying re: point drift but I don't think it would affect him as much since he'd be playing every day or so. And it seems incosistent/difficult to calculate.

But if you want to withdraw, that's fine. Maybe when he can play again he will let us know.

Cool. I think that's for the best. Sorry for the nastiness in my last post. I do get a little carried away sometimes.

[hsiale]

153 hsiale_ 5503 (+49)

After going 4-0 today finally I'm over 5500 rating