New iOS app shows real-time EVs as you play

[tomker]

Hi guys, I wanted to let you know about a new blackjack app that I made for iOS called "Blackjack X." It has a bunch of features for enthusiasts which I think a lot of you might be interested in:

- First, the big one, it displays EVs for each action as you're playing. (There'a a screen shot of this in the App Store.) To my knowledge, my app is the only piece of software that will do this, and the numbers are super interesting, or at least I think so.
- It also has a "hand analyzer" where you can quickly type in a hand and see EVs for all the possible actions. You can also specify a high-low count so you can see the EVs for that hand vs. a given count.
- I have calculated basic strategy for 6700+ rule variations (which took over 12 hours of cloud computing time) and include that database with the app, so you can change the rules in the app and immediately see the correct basic strategy for that rule set (and export it via email or print it from your device, if you want).
- It has a bunch of other features related to learning basic strategy which I assume would be less interesting to anyone in this forum, although I think it makes the app good for people you might know who aren't blackjack experts (e.g., a "training mode" which alerts you when you make a decision that isn't basic strategy and lets you cancel).

Let me know if you have any questions or feedback about the app. I would be happy to discuss!

[DSchles]

What method did you use for calculating the e.v. of any holding at any Hi-Lo TC? Representative decks?

Don

[tomker]

What method did you use for calculating the e.v. of any holding at any Hi-Lo TC? Representative decks?

Don

Unfortunately I don't think this can be done exactly in a convenient amount of time, so I do Monte Carlo simulation. For each hand of the simulation, I start with a full shoe and then randomly remove the appropriate number of high or low cards to hit the desired hi-lo count.

This process slows down the simulation, but even so, I find that the EVs converge pretty quickly and I think that they are off by only maybe ~0.01 after a couple seconds. Usually that's enough to see a clear difference between decisions. The simulation is run for 10 seconds though just to make sure. You can see the numbers change around as the simulation runs so you can get a feel for how fast they're converging.

[tomker]

Oh, BTW, as a follow up: I assume you were referring to the "hand analyzer" feature.

If you are talking about the "actual shoe" real-time odds, these EVs *are* calculated exactly based on the exact contents of the shoe and optimal play. All the necessary probabilities are enumerated and summed in real-time, except for a few cases where it's computationally prohibitive (basically splitting).

[Norm]

Unfortunately I don't think this can be done exactly in a convenient amount of time, so I do Monte Carlo simulation.

Unfortunately, representative subsets is not a valid methodology and orders of magnitude of more hands most be run.

The method used by CV uses massive tables built with nearly a trillion hands run on a overclocked PC with the voltage increased 19%. The methodology includes support for balanced counting systems and gives the dollars lost on an incorrect play. I still warn that it is only an estimate.

[tomker]

Unfortunately, representative subsets is not a valid methodology and orders of magnitude of more hands most be run.

Hmm, I thought I understood what was meant by "representative decks" but now I'm not sure. I assumed it meant picking a particular deck/shoe configuration that somehow ideally represented a particular hi-lo count and then doing calculations just based on that deck.

That is not what I'm doing though, since I agree that the result would be flawed, since in the real world you could have a bunch of different deck configurations that give you the same high-low but I'm sure would result in meaningfully different EVs in some cases.

So what I'm doing is generating a new random deck with the correct high-low count before each simulated hand. I believe this is mathematically valid.

Of course more hands is better when doing Monte Carlo but a few hundred thousand seem to give pretty accurate numbers, like I said, accurate to at least 1 cent per dollar. By looking at the EVs for the different decisions on the "Analysis" screen you can very easily see how much money you would lose per dollar bet due to an incorrect decision.

[tomker]

... The method used by CV uses massive tables built with nearly a trillion hands run on a overclocked PC with the voltage increased 19%. The methodology includes support for balanced counting systems and gives the dollars lost on an incorrect play. I still warn that it is only an estimate.

Oh, BTW, can I assume that you are involved in the development of CV since you know so much about how it works and was developed?

I would suggest checking out "the cloud" if you need to do a lot of computation. It might be a lot faster and more cost effective than running calculations on an overclocked personal computer.

In my case, I calculated my basic strategy database using a 16-core machine rented on a per-hour basis from DigitalOcean. It only cost a few dollars and took 1/4 of the time it would have taken on my relatively fast home computer. Very convenient and cost effective and kinda fun.

[Three]

Norm is the developer of CV.

[Norm]

1. "Cloud" is an advertising term, not a technology term. It has no specific technology meaning. It means whatever the advertiser claims it means. Kind of like "organic".
2. “Cloud” services nearly always use very slow processors, like the AMD sixteen core chips.
3. Representative decks generally means multiple representative decks, which your description appears to match. It is a shortcut used by “simulators” and does not present good results. I tried this technique a couple decades ago. I decided that it was not accurate. Indeed, there really is no reason that it should be considered accurate.
4. You claim to have used a DigitalOcean 16 core PC. I hope you realize that you are only using one core, and that core is very slow as 16 core PCs have slow cores.
5. A few hundred thousand hands as a sample is completely, absolutely useless. With eight million hands, the standard error would be 0.13. What do you think it is with a few hundred thousand? Even if you did a REAL simulation, instead of using an invalid technique -- in a few hundred thousand hand sim, you can actually prove that basic strategy in an H17 game has positive expectation.

Sorry if this seems harsh. But, you need to do some reading.

[Three]

Schooled!!!

[tomker]

Sorry Norm, I seem to have inadvertently angered you which was not my intention. I have not used CV but I have seen it referenced often by reputable sources and it seems like excellent software. I have a lot of respect for it (and thus you) and whatever I said that made you upset, please be assured it wasn't intended.

To answer a couple of your points:

1. You seem to have a very negative idea of cloud computing which I'm sure is justified in many cases, but not necessarily all cases. In fact I did rent a 16-core DigitalOcean computer for several hours. It ended up having a late-model 2.3GHz Intel Xeon processor... definitely not the fastest single-thread performer but multiply by 16 and it's pretty formidable. I definitely did use all 16 cores since my software ran exactly 16 times faster with 16 threads. I was also pleased to see that they have an effective cooling solution for their computers as I experienced no thermal throttling when all 16 cores were under full load. Overall I was very pleased with the experience and would happily recommend it to anyone needing to do a lot of computation.

3. I defer to your experience on the subject and techniques, but can you expand on why "representative decks" do not yield accurate results?

5. I'm not sure what you mean by standard error but I disagree that a few hundred thousand hands is "useless." If you re-run the simulations multiple times, you keep getting the same EVs, plus or minus ~0.005 and usually less. That indicates to me that the EVs are accurate to (conservatively) 1 cent per dollar. Obviously more iterations would be better but this seems like a useful amount of accuracy to me.

[Norm]

1. I am not the slightest bit angered or upset and gave no such indication. The Internet is, at times, difficult as you cannot hear a person’s voice or see his/her face. Try to avoid making assumptions about another’s attitude without clear verbal cues. I gave no such clues.
2. I did not give any view at all of any particular cloud computing site. I merely tried to inform you that it is an advertising term, and therefore has unclear meaning.
3. Representative decks limit the quintillions of possible samples to a relatively tiny number. Why would you do that?
4. If you rerun the sims with a few hundred thousand hands and get results that close, you have proved without any doubt that the sims are incorrect as they are far closer than standard error would dictate. Of course, this is one of the problems that occur with representative subsets.
5. If you do not know what “standard error” means, why are you posting on this subject at all?

Again, sorry if this sounds harsh. But, there are an embarrassingly large number of blackjack books based on ridiculously small samples proving all kinds of utter nonsense.

[Norm]

It ended up having a late-model 2.3GHz Intel Xeon processor... definitely not the fastest single-thread performer but multiply by 16 and it's pretty formidable.

Incidentally, there is no such processor. And if there were, it would not run at 16 times the speed.