PDA

View Full Version : How to test how often Set Mining works



Basic
01-14-2010, 01:30 PM
I think I must be making some sort of mistake when trying to calculate the success of my mining attempts using HEM, either that or I am very very very unlucky :eek:
I thought if I counted how many low pairs I have that seen the flop by using a filter and then count how many of these have sets in the flop I could get a more accurate view of whether I am running good, bad or average for these.


So far in my database I have 33028 hands recorded at this level and to check the success rate I tried using the filters as follows.

First completely clearing all filters.

Unfiltered: No. of hands = 33028
Applying the "Saw Flop = True" filter : No. of hands = 3415

keeping this set and:

Adding a "Hole Card" filter of the pairs 22,33,44...,88 gives
No. of hands = 915 (I think this means I have 915 games in total were I had low/med pairs and went onto see a flop)

keeping these set and:

Now adding a "Made Hand Values"| "Any Flop Hand Value" filter of "Three of a kind - set - High" and the "Second set" and the "Low set" sets gives
No. of hands = 49 (I think this means I have 49 hands that saw the flop and 1 card matched my pair (it may not count 4 of a kind, and may count hands were the flop was Y,Y,Y but this would be close enough for me)

This seems to indicate that I have hit sets only 49 times in 915 ie, 1 in 18.7 times! Yipes.

In theory I think I should fail to hit a set once in (48/50 + 47/49 + 46/48) = 0.88.
So I should hit approx. 12% of times which translates to the well known 1 in 7.5.


I think I must have used HEM wrong here because I haven't broken that many mirrors.
I stopped looking for leaks and tried to work out how often a 1 in 18 result should occur in theory. Sadly my maths was not up to the job but I did knock together a simple python script simulation and using this 1 in 18 does not occur in a 915 sample size even 1 in 1000000 times.

As I stated at the top I think I am screwing up the filters of HEM.

Has anybody got any clue as to what I am doing wrong.

Edit[ I now realize I also forgot about the time I flop full houses - but this will only add a few more hands so I still seem to be hitting at a rate of about 1 in 16 or so and that is so unlikely. It is still the case that I am missing the obvious and being a complete numbskull.) ]

Thanks for any help,
Basic
(and I apologize to the Poker Gods if I have somehow upset them :-)

netsrak
01-15-2010, 05:08 AM
You can't do such deep interpretations for a sample size of 33k hands.

Your filters look fine for me by the way.

Basic
01-15-2010, 06:40 PM
Thanks very much for the reply netsrak.



You can't do such deep interpretations for a sample size of 33k hands.

I do think tha 33k hands is low for any analysis - and this was actually one of the reasons that I wanted to know how successful the mining had been. If I had been hitting loads then I should earn loads from small pairs. The smaller the sample size the more important the success rate is to decide how well I am playing.

After doing a few searches through the forums I found that there is a selectable stat you can add to reports which is actually what I want - Flop Set%.
Using this reports the my success is 10.9% - slightly low but around what I would expect - unlike the 1 in 18 - 5.5% that I get when I use the filteres to try to acheive the same thing. Something seems a bit fishy - lol.

I am confused as to why I do not get the same(or even near) results when using the filters.

The sql for the the "Flop set%" includes a part:
FLOP.MadeHandValue in (60,61,62,120,140)

this seems to correspond to my selecting the filter "Hand Values"
"Three of a kind - set - High"
"Second set"
"Low set"
"Full House - Poket Pair - No Trips on board"
"Full House - Poket Pair + Trips on board"

It does seems to me that one of the filters is producing the wrong data.

And incidently I think ther is a small error in the "Flop set%" sql as this does not take into account the time when the board is trips eg, Y,Y,Y
Edit [ I think the "Full House - Poket Pair + Trips on board" should not be used as this gives X,X, - Y,Y,Y where you have not really hit your set]


Would it be possible for you to try the same approach with a larger data set - see if you can get the filters to agree with the value "Flop set%" reports.

Many thanks for you help and time,
BaseMetal

B-Money
01-16-2010, 04:45 PM
Basic,

Awesome you are using the deeper features inside of Holdem Manager. Many users don't.

99.9% of the time when a single user says our stats are calculated wrong, they don't understand how the stat is calculated. For your Flop Set %. Cool you found the Flop set %. Makes life easier for a %. :-)

So, to test this or any other stat. Filter for it like you are doing and replay the hands to make sure what is supposed to happen is happening. You mentioned board sets like 555 do not count. Correct, because everyone flopped this hand. Add the Hand Value, "Trips on Board." Save the filter and go back to the reports. You'll see the flop set % is NA. Which means it's not included in the stat.

You can do this with just about any stat inside of Holdem Manager. Instead of trying to look at code to figure out how stats are calculated, use the filters and double check it yourself. If you think a stat is calculated wrong, replay all the hands that come back for that situation to double check.

I hope this helps and GL,
-B

Basic
01-16-2010, 09:16 PM
Thanks B-Money



Instead of trying to look at code to figure out how stats are calculated, use the filters and double check it yourself.


I have been trying to do this - you are right about the "Flop Set %" it does ignore the X,X Y,Y,Y type hands correctly. I am still puzzled as to why I dont get the expected result from the filters - I am probably making some sort of mistake though.
I can't go through all the hands it will take too long - it's what the filters are for!

I will try to see the problem by starting with very few hands and keep adding until I get some unexpected result - but at the moment I've just got in from the pub so I'll hit the tables first - just joking

Anyway, I have in the past used postgres and sql on the HEM database but it takes an awful long time to work out want the fields are for eg, in the part
FLOP.MadeHandValue in (60,61,62,120,140)
I have to guess what "60,61,62,120,140" actually do - comparing/guessing against the Filter info.

I don't think you have published what the tables are - PT does and makes it easy to do sql. In the past I have used sql on the HEM data but I have had to run some report and record the sql log and infer want the fields are. Is there is another way of finding this out? I have tried using pgAdmin but the tables are just too complicated and I have failed. Can you help and add some explanations of the tables? The post:
Formulas for every HM Stat! (http://http://forums.holdemmanager.com/showthread.php?t=3595)

is very useful but an explanation of the tables would make life easy. If you have already posted an explanation could you provide a link - I cannot find the info.

Thanks for any help
Basic

B-Money
01-17-2010, 06:09 PM
If the tables are over your head in that forum I'd stick to Holdem Manager. It's complicated math that my education didn't provide either. :-)

I can't stress enough that you don't need to do this deep mathematical analysis with PostgreSQL. Holdem Manager will do it for you. That forum post you mention is for people that want to make their own stats. IE, smart programmer types.

Oh, you might want to check out the stat definition list we have too. It's a little outdated with new stats, but it's 1 of the things high on my list of things to update.

http://www.holdemmanager.net/faq/afmviewtopic.aspx?topicid=15


GL,
-B