HU NLHE: Bot beats pros(?) for 49 +/- 4 bb/100 - Page 3 - Poker News

Two Plus Two Forums Poker News & Discussion News, Views, and Gossip

HU NLHE: Bot beats pros(?) for 49 +/- 4 bb/100

Post Reply Subscribe

...

Page 3 of 6

1 2 3 4 5 6

Page 3 of 6

1 2 3 4 5 6

01-13-2017 , 07:27 AM

#51

vrael111

veteran

Join Date: Dec 2013 Posts: 2,949

Damn wrong thread

Quote

01-14-2017 , 10:06 PM

#52

samooth

veteran

Join Date: May 2009 Posts: 3,350

Quote:

Originally Posted by vrael111

Damn wrong thread

Truth.

Quote:

Originally Posted by vrael111

l1) why did they "tune" the bot for 200bb instead of 100bb where 99% of action is happening and where all the top pros excel at?

Here and here and here.

Quote

01-15-2017 , 05:52 AM

#53

amago

newbie

Join Date: Jun 2008 Posts: 20

what puzzles me is why didn't test against acpc bots Tartanian and theire own Act1. May be it's results against bots are not that good.

Quote

01-15-2017 , 10:13 AM

#54

samooth

veteran

Join Date: May 2009 Posts: 3,350

http://www.theverge.com/2017/1/10/14...-texas-hold-em

Let the marketing wars begin

Bot vs bot hu4rollz, do it!

Spoiler:

Quote

01-15-2017 , 11:35 PM

#55

that_pope

Carpal \'Tunnel

Join Date: Dec 2006 Posts: 33,226

Looks like a statistical tie from where I am sitting.

Last edited by that_pope; 01-15-2017 at 11:35 PM. Reason: Seriously, you guys are slow.

Quote

01-16-2017 , 11:43 AM

#56

ishipkq

old hand

Join Date: May 2012 Posts: 1,479

i once heard that otbs real name is mike phan tho

Quote

01-16-2017 , 11:59 AM

#57

samooth

veteran

Join Date: May 2009 Posts: 3,350

page 2:

AI poker researchers seem to love making comparisons between "HUNL" and go. This paper explains how the size of a texas holdem hu poker game is calculated -- some info:

- hu limit has ~1.38x10^13 info sets (all numbers account for isomorphisms)

- the first hunl format (2007-2008) in the ACPC was 1-2 with 1000 stacks (500bb), which has ~7.23x10^72 info sets

- then the format switched to 1-2 with 400 stacks (200bb) in 2009, which has ~1.39x10^48 info sets

- finally, magic happens:

resulting in ~6.37x10^161 (!) info sets

So now the narrative is that "HUNL" has close to the same size of go, and since it involves incomlete info while go does not, it must be at least equally as tough to solve if not tougher. One wonders how tough HUNL must be if you'd allow for betting cents instead of only whole dollar amounts! #bull****science

Last edited by samooth; 01-16-2017 at 12:04 PM.

Quote

01-18-2017 , 09:44 AM

#58

samooth

veteran

Join Date: May 2009 Posts: 3,350

How to market a statistical tie from the other side of the fence:

"huge..." quote is from pokerfuse

Quote

01-19-2017 , 06:54 AM

#59

punter11235

Carpal \'Tunnel

Join Date: Mar 2005 Posts: 8,210

I haven't followed that much but I've just read the paper about Deep Stack and I think it's great. Clearly written, all the terms are explained. Organization is great (important stuff at the top in human readable language, technical details in appendixes). Great job, it's really refreshing to see.

I also think the approach itself is promising although I already have some ideas to improve it but it has to wait

Again, it's very refreshing to see something which actually "thinks" during play and not only relies on running a super-computer for X months to make a huge table which is then read during play with little to no modifications.

Quote

01-19-2017 , 04:23 PM

#60

michaelph23

stranger

Join Date: Aug 2013 Posts: 6

OTR is polar vs cappded situation, so going all in maximizes EV. Pretty easy stuff. And he has to call 1-a.

Quote

01-19-2017 , 05:20 PM

#61

Abbaddabba

Pooh-Bah

Join Date: Jul 2005 Posts: 5,807

If you could bet 10x or 20x the pot you could bet that much too, but do you really think it would be significantly better? It just means that the bot calls with roughly half or 1/4 the frequency that it would against a 5x pot bet.

Last edited by Abbaddabba; 01-19-2017 at 05:26 PM.

Quote

01-28-2017 , 08:02 AM

#62

samooth

veteran

Join Date: May 2009 Posts: 3,350

Question 3:

bang bang. DO IT!

Quote

01-28-2017 , 04:07 PM

#63

tuccotrading

old hand

Join Date: Oct 2006 Posts: 1,373

Why did Victor Santos (OtB_RedBaron)
not play more hands against the bot?

Quote

03-02-2017 , 08:55 PM

#64

ArtyMcFly

Carpal \'Tunnel

Join Date: Dec 2014 Posts: 13,256

The makers of Deepstack have started some sort of publicity blitz to coincide with the release of their new research paper.

(Click the tweet above for a 30-message thread).

Unlike Libratus, it can play any stack size. It runs on a laptop, takes a maximum of 5 seconds to compute the best move and will apparently soon be destroying all-comers in HU tournaments on Twitch.

EDIT: Nerdy stuff at http://www.deepstack.ai

Last edited by ArtyMcFly; 03-02-2017 at 09:01 PM.

Quote

03-02-2017 , 09:33 PM

#65

AltruisticRaven

newbie

Join Date: Jul 2016 Posts: 48

It's so aggravating how poker AI market themselves. Maybe if they were actually doing something useful and novel (instead of running old algorithms on super computers) they would get the amount of attention they desire...

Quote

03-03-2017 , 02:57 AM

#66

florentinopeces

old hand

Join Date: May 2015 Posts: 1,199

Does anyone seriously think that the good bots are marketing themselves out in the open for all to see?

Quote

03-03-2017 , 03:36 AM

#67

ArronNoak

centurion

Join Date: Feb 2017 Posts: 142

Quote:

Originally Posted by florentinopeces

Does anyone seriously think that the good bots are marketing themselves out in the open for all to see?

yes

Quote

03-03-2017 , 04:14 AM

#68

NerdSuperfly

veteran

Join Date: Nov 2010 Posts: 2,172

Quote:

Originally Posted by AltruisticRaven

Quote:

Originally Posted by florentinopeces

Does anyone seriously think that the good bots are marketing themselves out in the open for all to see?

what you don't get, these AI attempts aren't just for 'poker'. game theory can applied on various scientific fields which includes not only (the obvious guess) finances and markets, but also different sectors, e.g. social studies, medicine ...

so of course universities try to get attention, so they get additional funds from new sources ... but okay, for most of the people living in 2+2 land it's just about lol-poker and how this affects mass tabeling at NLsomething

Quote

03-03-2017 , 04:48 AM

#69

AltruisticRaven

newbie

Join Date: Jul 2016 Posts: 48

Quote:

Originally Posted by NerdSuperfly

I understand that getting research grants / attention is super important for these groups, and I also understand that game theory research is valued in other fields outside poker...

But I don't understand what about both of the quoted posts implied that we didn't understand this. I'm simply stating that both Libratus and Deepstacked (but especially Libratus), aren't using any novel techniques to further optimize their poker bots. Optimizing their poker bots further without developing new techniques, does little or nothing to further advance research into game theory in other much more important areas.

And it's annoying to see them try and market themselves to be something that they aren't using their flawed / irrelevant human vs machine matches, and traditional methods of creating poker bots. Though I understand why they do it.

Last edited by AltruisticRaven; 03-03-2017 at 05:02 AM.

Quote

03-03-2017 , 05:48 AM

#70

florentinopeces

old hand

Join Date: May 2015 Posts: 1,199

The operative word in my post was "good".

Somebody explain to me what's the incentive to publicize a strong poker bot.

Serious strong AI research groups do not appear to be dedicating their time to poker. What little gets published seems old hat and run of the mill.

Quote

03-03-2017 , 03:19 PM

#71

nburch

newbie

Join Date: Jan 2015 Posts: 47

Short answer of why the research isn't run of the mill, and why "paradigm shift" is not totally overselling things.

Algorithms for imperfect information games like poker have always had to look at the whole game. That's a real downer, when you have a game with 10^160 decision points. It'll take forever, and you won't be able to write down what you learned because there aren't enough atoms in the universe.

Enormous size is clearly not as much of a problem for perfect information games, like go with 10^170 decision points. Computers have been good at chess (10^40?) for decades. Why? Because you can independently reason about different parts of the game. Figure out a way to combine this with a heuristic evaluation function -- a way to estimate what happens if you play out a situation with perfect play -- and you're set.

We introduced a theoretically sound way of reasoning independently about different parts of imperfect information games, and an effective way of combining it with a heuristic evaluation function. Something like the difference between staring at every single possible chess state, and using alpha beta. Very much NOT "old hat and run of the mill."

Quote

03-04-2017 , 01:23 AM

#72

skario

enthusiast

Join Date: Jan 2007 Posts: 87

Quote:

Originally Posted by nburch

This is the most important new thing in a long time.

Libratus is cool too obv, but I think this is way more significant.

Quote

03-04-2017 , 04:21 AM

#73

morilka

enthusiast

Join Date: Jan 2009 Posts: 84

Quote:

Originally Posted by ArtyMcFly

The makers of Deepstack have started some sort of publicity blitz to coincide with the release of their new research paper.

They posted HHs and there are A LOT of questionable hands:

Spoiler:

PokerStars - $100 NL (2 max) - Holdem - 2 players
Hand converted by PokerTracker 4

schwab.sebastian (BB): $20,000.00
DeepStack (SB): $20,000.00

DeepStack posts SB $50.00, schwab.sebastian posts BB $100.00

Pre Flop: (pot: $150.00) schwab.sebastian has 2

Pre Flop: (pot: $150.00) DeepStack has 8

DeepStack raises to $200.00, schwab.sebastian calls $100.00

Flop: ($400.00, 2 players) 9

schwab.sebastian checks, DeepStack raises to $200.00, schwab.sebastian calls $200.00

Turn: ($800.00, 2 players) J

schwab.sebastian bets $600.00, DeepStack raises to $1,600.00, schwab.sebastian raises to $4,600.00, DeepStack raises to $9,600.00, schwab.sebastian calls $5,000.00

River: ($20,000.00, 2 players) J

schwab.sebastian checks, DeepStack raises to $10,000.00 and is all-in, schwab.sebastian calls $10,000.00 and is all-in

DeepStack shows 8

(Full House, Jacks full of Twos)

schwab.sebastian shows 2

(Full House, Twos full of Jacks)

DeepStack wins $40,000.00

---

PokerStars - $100 NL (2 max) - Holdem - 2 players
Hand converted by PokerTracker 4

qin.youwei (BB): $20,000.00
DeepStack (SB): $20,000.00

DeepStack posts SB $50.00, qin.youwei posts BB $100.00

Pre Flop: (pot: $150.00) qin.youwei has 7

Pre Flop: (pot: $150.00) DeepStack has 4

DeepStack raises to $300.00, qin.youwei raises to $1,500.00, DeepStack calls $1,200.00

Flop: ($3,000.00, 2 players) J

qin.youwei bets $3,000.00, DeepStack calls $3,000.00

Turn: ($9,000.00, 2 players) K

qin.youwei bets $9,000.00, DeepStack raises to $15,500.00 and is all-in, qin.youwei calls $6,500.00 and is all-in

River: ($40,000.00, 2 players) 8

DeepStack shows 4

(Three of a Kind, Threes)

qin.youwei shows 7

(One Pair, Threes)

DeepStack wins $40,000.00

---

PokerStars - $100 NL (2 max) - Holdem - 2 players
Hand converted by PokerTracker 4

dmit.pol (SB): $20,000.00
DeepStack (BB): $20,000.00

dmit.pol posts SB $50.00, DeepStack posts BB $100.00

Pre Flop: (pot: $150.00) dmit.pol has 4

Pre Flop: (pot: $150.00) DeepStack has Q

dmit.pol calls $50.00, DeepStack raises to $20,000.00 and is all-in, dmit.pol calls $19,900.00 and is all-in

Flop: ($40,000.00, 2 players) 4

Turn: ($40,000.00, 2 players) 8

River: ($40,000.00, 2 players) 2

DeepStack shows Q

(Two Pair, Eights and Fours)

dmit.pol shows 4

(Two Pair, Eights and Fours)

DeepStack wins $40,000.00

---

PokerStars - $100 NL (2 max) - Holdem - 2 players
Hand converted by PokerTracker 4

bachmann.juergen (SB): $20,000.00
DeepStack (BB): $20,000.00

bachmann.juergen posts SB $50.00, DeepStack posts BB $100.00

Pre Flop: (pot: $150.00) bachmann.juergen has 4

Pre Flop: (pot: $150.00) DeepStack has A

bachmann.juergen raises to $200.00, DeepStack calls $100.00

Flop: ($400.00, 2 players) A

DeepStack checks, bachmann.juergen raises to $800.00, DeepStack raises to $1,800.00, bachmann.juergen calls $1,000.00

Turn: ($4,000.00, 2 players) T

DeepStack checks, bachmann.juergen raises to $18,000.00 and is all-in, DeepStack calls $18,000.00 and is all-in

River: ($40,000.00, 2 players) T

bachmann.juergen shows 4

(Flush, Ace High)

DeepStack shows A

(Full House, Aces full of Tens)

DeepStack wins $40,000.00

---

PokerStars - $100 NL (2 max) - Holdem - 2 players
Hand converted by PokerTracker 4

zurr.shai (BB): $20,000.00
DeepStack (SB): $20,000.00

DeepStack posts SB $50.00, zurr.shai posts BB $100.00

Pre Flop: (pot: $150.00) zurr.shai has 5

Pre Flop: (pot: $150.00) DeepStack has K

DeepStack raises to $200.00, zurr.shai raises to $800.00, DeepStack calls $600.00

Flop: ($1,600.00, 2 players) K

zurr.shai bets $1,200.00, DeepStack calls $1,200.00

Turn: ($4,000.00, 2 players) 4

zurr.shai bets $3,600.00, DeepStack calls $3,600.00

River: ($11,200.00, 2 players) 7

zurr.shai bets $7,400.00, DeepStack raises to $14,400.00 and is all-in, fold

DeepStack wins $26,000.00

---

also
1) AI played 20709 SBs and 16649 BBs, difference is quite significant.
2) AI has a casual 61% WSD (65% on BB and 57% on SB)
3) AI has very low VPIP on BB - 61,5%, may be due to a lot of limping from humans, but also may be because its actually bad.
4) given 1,2&3 some hands might've been deleted.

Quote

03-04-2017 , 04:25 AM

#74

ArronNoak

centurion

Join Date: Feb 2017 Posts: 142

Quote:

Originally Posted by morilka

...

also
1) AI played 20709 SBs and 16649 BBs, difference is quite significant.
2) AI has a casual 61% WSD (65% on BB and 57% on SB)
3) AI has very low VPIP on BB - 61,5%, may be due to a lot of limping from humans, but also may be because its actually bad.
4) given 1,2&3 some hands might've been deleted.

lol if tru

Quote

03-04-2017 , 12:42 PM

#75

PrimordialAA

Carpal \'Tunnel

Join Date: May 2007 Posts: 11,673

Quote:

Originally Posted by morilka

They posted HHs and there are A LOT of questionable hands:

Spoiler:

Pre Flop: (pot: $150.00) DeepStack has 8

DeepStack raises to $200.00, schwab.sebastian calls $100.00

Flop: ($400.00, 2 players) 9

schwab.sebastian checks, DeepStack raises to $200.00, schwab.sebastian calls $200.00

Turn: ($800.00, 2 players) J

schwab.sebastian checks, DeepStack raises to $10,000.00 and is all-in, schwab.sebastian calls $10,000.00 and is all-in

DeepStack shows 8

(Full House, Jacks full of Twos)

schwab.sebastian shows 2

Pre Flop: (pot: $150.00) DeepStack has 4

DeepStack raises to $300.00, qin.youwei raises to $1,500.00, DeepStack calls $1,200.00

Flop: ($3,000.00, 2 players) J

qin.youwei bets $3,000.00, DeepStack calls $3,000.00

Turn: ($9,000.00, 2 players) K

qin.youwei bets $9,000.00, DeepStack raises to $15,500.00 and is all-in, qin.youwei calls $6,500.00 and is all-in

River: ($40,000.00, 2 players) 8

DeepStack shows 4

(Three of a Kind, Threes)

qin.youwei shows 7

Pre Flop: (pot: $150.00) DeepStack has Q

dmit.pol calls $50.00, DeepStack raises to $20,000.00 and is all-in, dmit.pol calls $19,900.00 and is all-in

Flop: ($40,000.00, 2 players) 4

Turn: ($40,000.00, 2 players) 8

River: ($40,000.00, 2 players) 2

DeepStack shows Q

(Two Pair, Eights and Fours)

dmit.pol shows 4

Pre Flop: (pot: $150.00) DeepStack has A

bachmann.juergen raises to $200.00, DeepStack calls $100.00

Flop: ($400.00, 2 players) A

DeepStack checks, bachmann.juergen raises to $800.00, DeepStack raises to $1,800.00, bachmann.juergen calls $1,000.00

Turn: ($4,000.00, 2 players) T

DeepStack checks, bachmann.juergen raises to $18,000.00 and is all-in, DeepStack calls $18,000.00 and is all-in

River: ($40,000.00, 2 players) T

bachmann.juergen shows 4

(Flush, Ace High)

DeepStack shows A

Pre Flop: (pot: $150.00) DeepStack has K

DeepStack raises to $200.00, zurr.shai raises to $800.00, DeepStack calls $600.00

Flop: ($1,600.00, 2 players) K

zurr.shai bets $1,200.00, DeepStack calls $1,200.00

Turn: ($4,000.00, 2 players) 4

zurr.shai bets $3,600.00, DeepStack calls $3,600.00

River: ($11,200.00, 2 players) 7

zurr.shai bets $7,400.00, DeepStack raises to $14,400.00 and is all-in, fold

DeepStack wins $26,000.00

---

Where did you get these HHs? I might be accepting their challenge vs pros and would love to review them beforehand if possible

Quote

Page 3 of 6

First

1 2 3 4 5 6

Last

Post Reply Subscribe

...

Page 3 of 6

First

1 2 3 4 5 6

Last