Print

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

[Seprix]

Woah. Ela's in Dominion!

[Kirian]

[GeoLib]

So I've read the full poker paper. Here's my understanding of what's going on.

I actually met this guy last summer; he said he had it down to one milli-big blind off the nash equilibrium, but hadn't published yet. I guess now he has

Huh - my impression was that it wasn't the exact solution to the nash equilibirum, and was generated from a very, very large sample of actual game states of poker. Maybe I'm wrong though. Or, maybe they did it over all states, and worked on it til convergence. That way, it's not necessarily exactly perfect because of built-in limitations to how accurately the computer can represent decimals, but it's close enough to not even matter.

The "poker is solved" headline has been giving me mathematical twinges too.  Going on nothing else but the news coverage, this is roughly what I'm imagining has happened.

I've a vague idea I heard once that, in the same way that there's an algorithm that will provide a proof for any provable statement, there's an algorithm that is asymptotically optimal at playing games (or maybe it was trading, or perhaps there's no difference).  The idea is that you just try all of the possible strategies and measure how well they all do, and spend progressively more of your time using the better ones.  So what I imagine is that an algorithm like this has been run on this type of poker for a really long time, and that all of the clever work has gone into making it remotely computationally feasible.

Counterfactual Regret Minimization (CFR) has been used on poker for a while now. The innovations in this paper are to:
- store everything compressed on disk to deal with the storage problem
- use a variant called CFR+ that looks at the whole game tree at each iteration and constrains regrets to be non-negative. This means that actions that appeared to be bad and then some evidence suggested that they were actually good will immediately have positive regret (indicating some regret for having not played them) rather than being weighed down by a bunch of negative regret. So they will be entered into the pool of possible actions to choose immediately upon proving at all useful.
- Normally the strategy that is used is the average of all strategies used, but they observed that the exploitability of the current strategy tends towards zero, so they use that instead of the average.

In order to determine the exploitability of a strategy, they pit it against the worst-case opponent strategy. They were able to make this calculation computationally feasible by exploiting some details of the structure of the imperfect information game. That's how they got the <1/1000 of a big blind per game.

I feel like the article leaves out the most interesting part: how does the bot play?  How often does it bluff?  How often does it slow play?  How aggressive is it?  How important is position?  What percentage of hands does it play?  How often does it take the pot with the weaker hand?  And for me the most interesting question would be, if it plays against itself, what percentage of hands make it to showdown?

After Googling around, it seems like the original paper has some stats on how it plays - for instance, it called on the first action 0.06% of the time. The actual paper is going to be trickier to get access to.

I'm not sure if that's from a large sample of possible states, or explicitly from every possible state. For qualitative questions, it's very hard to quantify some of them, and it's entirely possible the bot pays wildly differently on states that appear to be similar in value, so at best you can hope that you can play the bot yourself somehow...

In the paper they put exact bounds on the value of the game at between 0.0877 and 0.0897 big blinds per game in favor of the dealer.

Some strategy details that they talk about are that the dealer almost never calls and usually raises. The strategy almost never raises all the way to the betting limit on the first round of betting. It folds on fewer hands than humans do (as non-dealer. It very rarely folds as dealer).

[SwitchedFromStarcraft]

....Some strategy details that they talk about are that the dealer almost never calls and usually raises. The strategy almost never raises all the way to the betting limit on the first round of betting. It folds on fewer hands than humans do (as non-dealer. It very rarely folds as dealer).

In heads up, the button (dealer) raising instead of calling has been a good strategy for a long time, nothing shocking there.

I can't get to the paper itself, so can you clarify your second comment? The phrase "almost never raising to the betting limit" is a little confusing.  In fixed limit holdem, any raise will be to the "betting limit", as the limits are fixed.  Do you mean that the strategy almost never puts in the final raise (in a structure allowing the original bet and say 3 raises)?  That would seem obvious, as the opponent would have to be putting in raises as well for anyone to have an option to put in the final raise - this is heads up after all. The opponent not raising would by obviate your ability to raise to the maximum.

[pacovf]

Can I ask for random tech support here?

I have a problem with my internet connection. I live in a student residence. At first the internet was free access, and there was no problem.

Then they added some control on the internet connection: when I open a browser, it automatically opens a login page where I have to identify myself before I can actually go on the internet. The problem is, it seems to have broken something, because I can't see embedded youtube images, I can't access half my google results (and can only access google image results when using google chrome instead of firefox), a lot of sites take forever to open (20-30 secs), and very often I have to manually type the site address with https:// in front to actually open a page (ex: I can't access wikipedia otherwise), and some random sites are completely blocked (particularly annoying when trying to download things from sourceforge, for example). I've complained to the internet provider, but they are not helping.

Now, today I've tried connecting with my work laptop, uses Ubuntu instead of Windows Vista (yes, I know, who uses Vista), and I have none of this problems.

So, does anyone have any idea what is the problem I have, and how to fix it?

[KingZog3]

Possibly it's this new internet security they put up. If they're using some service to make their internet secure then it may blocking too many things. I've been on connections that won't let me connect to f.ds because it's a "threat."

[pacovf]

But that isn't the problem. They aren't blocking the sites because they are a threat, I just cannot connect to these sites for some unexplained reason (the browsers try loading the site for a minute or so, and then it timeouts or something). If they were blocking the sites for security reasons, I wouldn't be able to access them by using another computer with the same internet connection.

[qmech]

I'm looking to buy my first pair of reasonable headphones, and some of you seem to know about this kind of thing.  I listen to very little music, and mostly want them for listening to soundscapes on youtube recorded with binaural microphones.  At the moment I'm using a £10 headset; I suspect I'm missing detail, and am prepared to spend some money to find out.

I'm aiming to spend around £50, but am looking for value rather than cheap.  My searching so far has led me to the Sony MDR7506 (£75) and the Sennheiser HD449 (£60).  Any thoughts?  One potential problem with the MDR7056 appears to be that they might provide extremely accurate reproduction of people's not-that-great microphones.

[KingZog3]

Actually I'm also interested in headphones. Preferably ones that are easy enough to wear while taking public transit to listen to music, and also for editing film stuff.

[Seprix]

Actually I'm also interested in headphones. Preferably ones that are easy enough to wear while taking public transit to listen to music, and also for editing film stuff.

Don't buy Skullcandy or Beats Audio. Those suck, and are dumb because they charge tons for something that is completely inferior. Think of Macbooks and their pricing vs. windows computers, except they aren't actually as useful as a Macintosh in functionality. The best high end headphones are usually Sony and Bose.

Spending $75 (50 pounds about) isn't going to cut it if you want full quality. Full quality costs about $500. But let me be clear, you don't need full quality to listen to great sound. So...

http://www.pcmag.com/article2/0,2817,2287680,00.asp

These are generally high rated headphones for about your price range.

[theory]

Do you want over-the-ear, on-ear, or in-ear?  That makes a significant difference.  For traveling I prefer in-ear earbuds and highly recommend the M9 (or M9P if you need a mic).

[qmech]

Do you want over-the-ear, on-ear, or in-ear?  That makes a significant difference.  For traveling I prefer in-ear earbuds and highly recommend the M9 (or M9P if you need a mic).

I think that's more directed at KingZog, but since I didn't state it explicitly I'm after over ear (with a closed back).

[SwitchedFromStarcraft]

I'm looking to buy my first pair of reasonable headphones, and some of you seem to know about this kind of thing.  I listen to very little music, and mostly want them for listening to soundscapes on youtube recorded with binaural microphones.  At the moment I'm using a £10 headset; I suspect I'm missing detail, and am prepared to spend some money to find out.

I'm aiming to spend around £50, but am looking for value rather than cheap.  My searching so far has led me to the Sony MDR7506 (£75) and the Sennheiser HD449 (£60).  Any thoughts?  One potential problem with the MDR7056 appears to be that they might provide extremely accurate reproduction of people's not-that-great microphones.

I'm familiar with both units.  There are lots of reviews online for each of these, so google is your friend.  Audio reviews are like wine reviews though (sound is subjective), so use them as a starting point, then arrange a demo of each at all possible.  The good news is that your source material is accessible from your phone or tablet, or an in-store computer.

A few general thoughts:  a) I think you are doing the right thing by looking at around-the-ear models.  b) With around-the-ear phones, how comfortable they are during extended wearing is very nearly as important as how they sound (unless you only have them on for 3 or 4 minutes at a time).  If you demo them, keep them on for at least 15 or 20 minutes (more on this in a moment).

In terms of features, the Sennheiser is going to be a little lighter in weight, but comes with a shorter cord. Sennheisers do not fold, the Sony pair does, if storage is an issue.  Both are wired, with the cord attaching only to the left earpiece.  That sometimes matters if you are moving around in a studio, at a mixing board, or similar.  Sitting in one spot plugged into your computer it would be less of a factor.  Ear pads on both are user replaceable and inexpensive.  (This will matter only if you keep the headphones a long time, and either of these will last you a VERY long time if not chewed on by dogs or thrown about by kids).

The Sonys have been around since 1991.  I own a pair of their precursor/older bother that is in the same price range in the US, the MDR-V6.  Both Sony models are well respected headphones designed many years ago (1985 and 1991).  They are very similar.  The V6 has just recently been discontinued by Sony, so you may begin seeing it at clearance prices in your area.  Full retail was about $100 here (67 pounds?).

Now I'm going to throw you a curve.  You don't say what your current headphones are, but I can suggest/strongly recommend a VERY comfortable, very capable around-the-ear headphone for about $25 (16 pounds?) - the JVC HA-RX500.  This is one of 3 pairs I own (the third pair being two orders of magnitude more expensive), and unless I am sitting down for extended critical listening, I reach for the JVC's first.  Why? Comfort.  They (arguably) don't have some of the strengths of the V6's, but they don't miss by much, and they are much more comfortable, so when I'm streaming Pandora for hours at a time while at my desk, I want them over the Sonys.  At that price point, their value is enormous, and I don't think you can go wrong.  They are almost guaranteed to be a step up from what you have, they will last a long time, and you'd get an answer to your question about source material issues without a huge investment.  And in the off chance that you didn't like the actual headphone (for reasons other than source material), you'd still have most of your budget in your pocket to try one of the models you enumerated.

TL:DR - You can't go wrong with either choice, but given your concern about source material, a third option in my final paragraph may be an attractive option.

Edit: To eliminate a double negative, and clarify one point.

[Ozle]

I'm looking to buy my first pair of reasonable headphones, and some of you seem to know about this kind of thing.  I listen to very little music, and mostly want them for listening to soundscapes on youtube recorded with binaural microphones.  At the moment I'm using a £10 headset; I suspect I'm missing detail, and am prepared to spend some money to find out.

I'm aiming to spend around £50, but am looking for value rather than cheap.  My searching so far has led me to the Sony MDR7506 (£75) and the Sennheiser HD449 (£60).  Any thoughts?  One potential problem with the MDR7056 appears to be that they might provide extremely accurate reproduction of people's not-that-great microphones.

I'm familiar with both units.  There are lots of reviews online for each of these, so google is your friend.  Audio reviews are like wine reviews though (sound is subjective), so use them as a starting point, then arrange a demo of each at all possible.  The good news is that your source material is accessible from your phone or tablet, or an in-store computer.

A few general thoughts:  a) I think you are doing the right thing by looking at around-the-ear models.  b) With around-the-ear phones, how comfortable they are during extended wearing is very nearly as important as how they sound (unless you only have them on for 3 or 4 minutes at a time).  If you demo them, keep them on for at least 15 or 20 minutes (more on this in a moment).

In terms of features, the Sennheiser is going to be a little lighter in weight, but comes with a shorter cord. Sennheisers do not fold, the Sony pair does, if storage is an issue.  Both are wired, with the cord attaching only to the left earpiece.  That sometimes matters if you are moving around in a studio, at a mixing board, or similar.  Sitting in one spot plugged into your computer it would be less of a factor.  Ear pads on both are user replaceable and inexpensive.  (This will matter only if you keep the headphones a long time, and either of these will last you a VERY long time if not chewed on by dogs or thrown about by kids).

The Sonys have been around since 1991.  I own a pair of their precursor/older bother that is in the same price range in the US, the MDR-V6.  Both Sony models are well respected headphones designed many years ago (1985 and 1991).  They are very similar.  The V6 has just recently been discontinued by Sony, so you may begin seeing it at clearance prices in your area.  Full retail was about $100 here (67 pounds?).

Now I'm going to throw you a curve.  You don't say what your current headphones are, but I can suggest/strongly recommend a VERY comfortable, very capable around-the-ear headphone for about $25 (16 pounds?) - the JVC HA-RX500.  This is one of 3 pairs I own (the third pair being two orders of magnitude more expensive), and unless I am sitting down for extended critical listening, I reach for the JVC's first.  Why? Comfort.  They (arguably) don't have some of the strengths of the V6's, but they don't miss by much, and they are much more comfortable, so when I'm streaming Pandora for hours at a time while at my desk, I want them over the Sonys.  At that price point, their value is enormous, and I don't think you can go wrong.  They are almost guaranteed to be a step up from what you have, they will last a long time, and you'd get an answer to your question about source material issues without a huge investment.  And in the off chance that you didn't like the actual headphone (for reasons other than source material), you'd still have most of your budget in your pocket to try one of the models you enumerated.

TL:DR - You can't go wrong with either choice, but given your concern about source material, a third option in my final paragraph may be an attractive option.

Edit: To eliminate a double negative, and clarify one point.

Tl dr Summary for everyone:   Just use the standard Apple ones

[pacovf]

Edit: To eliminate a double negative

[Awaclus]

Actually I'm also interested in headphones. Preferably ones that are easy enough to wear while taking public transit to listen to music, and also for editing film stuff.

Don't buy Skullcandy or Beats Audio. Those suck, and are dumb because they charge tons for something that is completely inferior. Think of Macbooks and their pricing vs. windows computers, except they aren't actually as useful as a Macintosh in functionality. The best high end headphones are usually Sony and Bose.

I agree about Skullcandy and Beats Audio, but Sony and Bose? Not AKG, Sennheiser, Beyerdynamic, Audio-Technica, Shure or Ultrasone?


I have positive experiences about Shure SE-215 earphones, which I mostly use for drumming. I've used them for editing film stuff and public transit and they're good for both purposes, especially the latter.

[qmech]

That's very useful, thank you.

Now I'm going to throw you a curve.  You don't say what your current headphones are, but I can suggest/strongly recommend a VERY comfortable, very capable around-the-ear headphone for about $25 (16 pounds?) - the JVC HA-RX500.  This is one of 3 pairs I own (the third pair being two orders of magnitude more expensive), and unless I am sitting down for extended critical listening, I reach for the JVC's first.  Why? Comfort.  They (arguably) don't have some of the strengths of the V6's, but they don't miss by much, and they are much more comfortable

I'm glad you mentioned these.  I've half-heartedly looked at getting some headphones once before, and ran across a similar very cheap and highly regarded set then (I think they were also JVC).  There happened to be a pair in a filing cabinet in my office (or perhaps that's where I first found the name) and I'd tried them a couple of times at work, but they had absolutely no padding across the top band and hurt quite a bit after half an hour or so.  If you say that these are comfortable then I've no reason not to buy a pair.

Thanks for taking the time to write such a helpful response.

[SwitchedFromStarcraft]

Tl dr Summary for everyone:   Just use the standard Apple ones

Nooooooooooooooooooo............

That would be like buying a toothbrush from Toyota, or a chainsaw from Merrill Lynch.


@qmech - You are most welcome, glad it helped.  These are indeed comfortable, and though inexpensive, are not at all cheap.  They represent great value, IMO.

[Ozle]

Tl dr Summary for everyone:   Just use the standard Apple ones

Nooooooooooooooooooo............

That would be like buying a toothbrush from Toyota, or a chainsaw from Merrill Lynch.


@qmech - You are most welcome, glad it helped.  These are indeed comfortable, and though inexpensive, are not at all cheap.  They represent great value, IMO.

Woah! Toyota do toothbrushes! I want one...

[qmech]

Tl dr Summary for everyone:   Just use the standard Apple ones

Nooooooooooooooooooo............

That would be like buying a toothbrush from Toyota, or a chainsaw from Merrill Lynch.

Woah! Toyota do toothbrushes! I want one...

Make sure you go for a diesel.  The electric ones are a menace; I never hear them coming.

[GeoLib]

....Some strategy details that they talk about are that the dealer almost never calls and usually raises. The strategy almost never raises all the way to the betting limit on the first round of betting. It folds on fewer hands than humans do (as non-dealer. It very rarely folds as dealer).

In heads up, the button (dealer) raising instead of calling has been a good strategy for a long time, nothing shocking there.

I can't get to the paper itself, so can you clarify your second comment? The phrase "almost never raising to the betting limit" is a little confusing.  In fixed limit holdem, any raise will be to the "betting limit", as the limits are fixed.  Do you mean that the strategy almost never puts in the final raise (in a structure allowing the original bet and say 3 raises)?  That would seem obvious, as the opponent would have to be putting in raises as well for anyone to have an option to put in the final raise - this is heads up after all. The opponent not raising would by obviate your ability to raise to the maximum.

I have no experience with high-level strategy, but the paper claimed that there was still disagreement about whether it was advisable to "limp" (i.e. call instead of raise) amongst human players. They do say that conventional wisdom is that limping is bad and then say that their algorithm supports said wisdom.

Yes, it doesn't make the final raise. I'll just quote the article "The strategy almost never “caps” (i.e., makes the final allowed raise) in the first round as the dealer, whereas some strong human players cap the betting with a wide range of hands."

[GeoLib]

Can I ask for random tech support here?

I have a problem with my internet connection. I live in a student residence. At first the internet was free access, and there was no problem.

Then they added some control on the internet connection: when I open a browser, it automatically opens a login page where I have to identify myself before I can actually go on the internet. The problem is, it seems to have broken something, because I can't see embedded youtube images, I can't access half my google results (and can only access google image results when using google chrome instead of firefox), a lot of sites take forever to open (20-30 secs), and very often I have to manually type the site address with https:// in front to actually open a page (ex: I can't access wikipedia otherwise), and some random sites are completely blocked (particularly annoying when trying to download things from sourceforge, for example). I've complained to the internet provider, but they are not helping.

Now, today I've tried connecting with my work laptop, uses Ubuntu instead of Windows Vista (yes, I know, who uses Vista), and I have none of this problems.

So, does anyone have any idea what is the problem I have, and how to fix it?

This sounds like something that you're not really going to be able to get help with except by complaining to your school. If sites are blocked you could try using a VPN like hola.

[pacovf]

I shouldn't have used the word "blocked", it throws people off. It is not my internet provider which is "blocking" these sites, it is some mix of their authentication program and my computer settings that breaks something down when trying to access some sites (an example being wikipedia), since I didn't encounter this problem before they set up the authentication, nor do I encounter it now when I use a different computer but the same internet connection (same plug, same login).

Thanks for helping anyway. I've tried complaining to my internet provider (repeatedly), but they just ignore me, probably because it is the society that manages the residence that choses the provider, not me...

[scott_pilgrim]

Yes, it doesn't make the final raise. I'll just quote the article "The strategy almost never “caps” (i.e., makes the final allowed raise) in the first round as the dealer, whereas some strong human players cap the betting with a wide range of hands."

Well, probably this is something that depends on who you're playing against.  Since the bot is trying to find the Nash equilibrium, it's going to assume that if the other player raises repeatedly, it most likely has a very strong hand (assuming that's the right thing to do and bluffing pre-flop is rare enough to be a negligible consideration).  But if you know your opponent, you can jam the pot to get some extra money in there for your big hands (if they're too loose), or to get them to fold against your weaker hands (if they're too tight).

[Kirian]

I'm looking to buy my first pair of reasonable headphones, and some of you seem to know about this kind of thing.  I listen to very little music, and mostly want them for listening to soundscapes on youtube recorded with binaural microphones.  At the moment I'm using a £10 headset; I suspect I'm missing detail, and am prepared to spend some money to find out.

I'm aiming to spend around £50, but am looking for value rather than cheap.  My searching so far has led me to the Sony MDR7506 (£75) and the Sennheiser HD449 (£60).  Any thoughts?  One potential problem with the MDR7056 appears to be that they might provide extremely accurate reproduction of people's not-that-great microphones.

Don't forget the $15000 gold-encrusted diamond-coated Monster cables, forged in Mt. Doom, then assembled on the ISS for special vacuum-proofing.