Disturbing Realization

MidnightFoxMidnightFox Red Chipper Posts: 321 ✭✭✭
edited August 2017 in Poker Software
@Doug Hull
Maybe I don't understand and you could show me, but it seems like the top number of each hand type is not X% of hands total, but X% of A-high/gutshot or better. So percentage above Top Pair, means, the percentage of A-high/Gutshot or better hands that are Top Pair. In other words total air hands are not factored into that top number.

r3w2botabqgm.jpg
(3000 flop sample)
Let's check those upper numbers on the left panel for T9s against Flopzilla
l8mtzq37et8i.jpg

Look at the discrepancy in Top Pair. Flop Falcon says 15%, but Flopzilla says 10.5%. But what happens when you take A-high or better (69.6% of hands), and you find out what percentage of hands of that is top pair...? (The 5 dead cards are simulating 3 flop cards and our opponents cards/this may not be correct)
10.5 divided by 69.6= 15.1 (Flop Falcon's figure)

Now look back at the Right column of Flop Falcon. No matter what range you enter, if you select all hand types as hits, both top and bottom percentages are identical:
iu0na3263gl2.jpg

If the top numbers are percentage of total hands, then the top and bottom percentages shouldn't be identical. There should be so much air (worse than A-high/gutshot), that the bottom number should be dramatically larger than the top number.

Also notice that you can add up all the top numbers on either column and it adds up to 100%; again, this should not be true. There should be a gap in the figures for air hands.

"...percent that is over Top Pair, is the overall chance that we hit Top Pair"
-Doug Hull, "2. Scenario Setup and Hit Requirements" video

:Jd:Tc

Comments

  • Doug HullDoug Hull RCP Coach Posts: 1,876 -
    Please give full screenshots when posting, there is a screenshot utility built into Flop Falcon. All the input to replicate a Flop Falcon scenario is captured in a screenshot, so it makes it easier to reproduce.

    I don't think I understood the first bit of your post.

    I will start with this:
    @Doug Hull

    Look at the discrepancy in Top Pair. Flop Falcon says 15%, but Flopzilla says 10.5%. But what happens when you take A-high or better (69.6% of hands), and you find out what percentage of hands of that is top pair...? (The 5 dead cards are simulating 3 flop cards and our opponents cards/this may not be correct)
    10.5 divided by 69.6= 15.1 (Flop Falcon's figure)

    I can not speak to how Flopzilla works. I can speak to how Flop Falcon does. Let's look at how a typical calculation goes:

    I set up the scenario I saw in your screenshot. Instead of hitting "1000 more flops" click on the three cards under the range selection charts. In this screenshot

    v4nw14oblmd7.png

    When you click those three cards the following happens:

    Random cards are dealt out until one that is in Hero's range is dealt. There are now 50 cards left in the deck.

    Random cards are dealt out until one in Villain's range is found. There are now 48 cards left in the deck.

    These are the two hands that will be studied for this iteration. :TD::9D: vs :3D::2S: in this case. There is already a critical difference from the Flopzilla screenshot shown. In the Flopzilla screenshot, I see that all deuces and half the treys are "dead" meaning they can not come out on the flop. So Flopzilla has a Hero range of exactly 9Ts and 44 cards left in the deck because the Deuces and half the Treys are dead.

    Immediately we see that the pieces of software have different initial conditions. Different initial conditions can and should lead to different results.

    Back to how Flop Falcon works. Out of this 48 card deck, a random flop is dealt. Because of the screenshot, I can not confirm is we are using filtered or unfiltered flops. I will presume unfiltered, so any random flop generated from the 48 card deck is counted. In this case the flop is: :9H::5C::4C:

    Notice the green dots that are on the screen. In this case, Hero's hand was classified as Top Pair (indicated by the green dot).

    This mode, of clicking flops one at a time allows you 100% transparency. You can see exactly how Flop Falcon classifies each hand and board combination.

    Flop Falcon places every iteration into exactly one classification. Using the custom tab, I have set up a filtered flop such that I will quickly get a hand where I have Top pair and a Flush draw.

    64lvd3db9r8l.png

    This is classified as a top pair in Flop Falcon.

    What does Flopzilla do? I do not know. Do they consider it a draw, do they consider it top pair, do they consider it both, do they consider it a combo draw? I do not have visibility into how they do things.

    We have found a situation that can be classified in different ways. Add on to this that the two pieces of software are have been set with different initial conditions. It should not be surprising there are different results.

    What I do know, is that Flop Falcon has complete transparency. By using custom tab you can force pretty much any flop and hand scenario you want very quickly. You then can see how Flop Falcon classifies that. As you run boards one by one, you can see the percentages change and easily keep track with pencil and paper to confirm the percentages are as expected.

    Co-founder Red Chip Poker,
    Author Poker Plays You Can Use
    Author Poker Workbook for Math Geeks
  • MidnightFoxMidnightFox Red Chipper Posts: 321 ✭✭✭
    edited August 2017
    Enter any range into Flopzilla and how much it should hit Top Pair. Then run 3000 hands on Flop Falcon with that range and the opposite range having a low impact on the range that is testing for Top Pair.
    hjtoqoufe2di.jpg
    xd38zzr1ac2l.jpg
    24% vs 19.8%

    Forget my Flopzilla fallacies. I'm searching to make sense of this. How do all the top numbers from each column equal 100% with so many air hands unaccounted for? Why is it, when you select all filters, all top and bottom numbers are identical? I think that's something you can reproduce for yourself despite my poor attempts. Is my software the only copy that is doing this? Do I just not understand?

  • Doug HullDoug Hull RCP Coach Posts: 1,876 -
    edited August 2017
    Enter any range into Flopzilla and how much it should hit Top Pair. Then run 3000 hands on Flop Falcon with that range and the opposite range having a low impact on the range that is testing for Top Pair.
    hjtoqoufe2di.jpg
    xd38zzr1ac2l.jpg
    24% vs 19.8%

    Flopzilla and Flop Falcon are giving different counts for Top Pair.

    I know how Flop Falcon is getting it's answer and you can follow Flop Falcon getting that answer by clicking through the flop area, seeing the classification of each and every flop.

    To resolve this discrepancy, you might want to ask the developers of Flopzilla how they arrive at their numbers. I did not develop Flopzilla, so I can only guess at what they are doing. They may have different classifications (around combo draws as discussed), or possibly be using mathematical shortcuts and look up tables that are faster but are approximations.
    Forget my Flopzilla fallacies.

    It seems you were satisfied with the classifications, so now want to understand this issue:
    I'm searching to make sense of this. How do all the top numbers from each column equal 100% with so many air hands unaccounted for? Why is it, when you select all filters, all top and bottom numbers are identical? I think that's something you can reproduce for yourself despite my poor attempts. Is my software the only copy that is doing this? Do I just not understand?

    How do the top numbers of each column add up to 100% meaning there is air that should not be classified into any bucket unaccounted for.

    I just dove into the code. There was an error where the percentage was calculated with the wrong denominator. This also had some effect on the issue discussed above. The difference in how hands are classified still stands, but this wrong denominator issue did effect it. This has been fixed. I have a patch release of Flop Falcon in the works. This bug fix will be in the next release.

    When the release is available, Flop Falcon will notice on startup and instruct you to download it.

    Thank you!
    Doug

    For the math inclined: The denominator to calculate the percentages is supposed to be the number of flops seen. Somewhere deep in the code when this final percentage was calculated. The sum of all the classifications was used. Since the airballs are not explicitly in the list of classifications, the sum of all classifications is NOT equal to the number of flops. This was a simple fix once noticed.
    Co-founder Red Chip Poker,
    Author Poker Plays You Can Use
    Author Poker Workbook for Math Geeks
  • tfaziotfazio Red Chipper Posts: 819 ✭✭✭
    Corner cases are to be expected with new software. I have been using Flop Falcon for some time and it works as expected.
  • Doug HullDoug Hull RCP Coach Posts: 1,876 -
    Version 1.47 has been released with this bug fix. The next you start Flop Falcon, it will ask you to download. Thank you for finding this error.
    Co-founder Red Chip Poker,
    Author Poker Plays You Can Use
    Author Poker Workbook for Math Geeks
  • Adam WheelerAdam Wheeler Red Chipper Posts: 2,659 ✭✭✭✭
    @Doug Hull

    What is the cliff notes about the errors ?
  • Doug HullDoug Hull RCP Coach Posts: 1,876 -
    Co-founder Red Chip Poker,
    Author Poker Plays You Can Use
    Author Poker Workbook for Math Geeks

Leave a Comment

BoldItalicStrikethroughOrdered listUnordered list
Emoji
Image
Align leftAlign centerAlign rightToggle HTML viewToggle full pageToggle lights
Drop image/file