Mentor Count

[zengrifter]

You look like a psychodelic crazed gremlin?

Psychedelic crazed GANESH.

Sir BJ, please run a near identical sim and include a third system - ZEN. Thanks. z(crazed one)g

 

[Myth]

Wouldn't the Mentor results be skewed in CVCX if you did not use the Custom TC feature? And match it to the book? Or I guess use full deck but cut the indexes in half?

 

[QFIT]

Custom count was used in the CVCX canned sims for Mentor. But again, twice as many indexes were used for the canned HiLo Full indexes sim as were used in the canned Mentor full indexes sim because ProBJ supplied twice as many indexes. The Sweet 16 sims are a better comparison, and they show Mentor outperforming level I systems.

 

[Myth]

Oh ok got ya, thought maybe the sims BJcount ran might have been different

 

[bjcount]

Psychedelic crazed GANESH.

Sir BJ, please run a near identical sim and include a third system - ZEN. Thanks. z(crazed one)g

ZG,
The first set (mentor vs HiLo DD & 6D) in the earlier post, I ran are on my laptop which I'll run a similar sim using zen for you tomorrow. The second set which is in this desktop is DD only, so I ran a sim using Zen for comparison purposes.

BJC


Ga·nesh (gə-nāsh') n. Hinduism
The god of wisdom and the remover of obstacles, son of Shiva and Parvati, depicted as a short fat man with an elephant's head.
Is that how you want us to know you?

 

[bjcount]

ZG,

Here are the sims from the CVData configured on my laptop. You will notice I added a sim for HiLo +/-10. Based on QFITS earlier response

twice as many indexes were used for the canned HiLo Full indexes sim as were used in the canned Mentor full indexes

I modified the tables so there were no TC's higher then +10 or lower then -10. For example, 16v5 is hit at TC< -13. I changed this to always stand.

BJC

 

[Renzey]

ZG,

Here are the sims from the CVData configured on my laptop. You will notice I added a sim for HiLo +/-10. Based on QFITS earlier response I modified the tables so there were no TC's higher then +10 or lower then -10. For example, 16v5 is hit at TC< -13. I changed this to always stand. BJC

Are your index numbers for Mentor being "scaled" appropriately? Notice that its published indices scale them according to the "count-per-two-decks" TC method. Hence, the index for 15 vs. 10 is +15 TC. If you were to run the sim using the conventional "count-per-deck" TC, this play should kick in at +7 or +8 TC. Furthermore, if you're limiting the index plays that get tripped in by using a fixed TC number, even after making the appropriate TC adjustments, certain Mentor index plays will be will be deleted by mere virtue of its higher card tags.

There has to be something glaringly wrong with the Mentor sim results, and I'm not sure what it is. I can think of two cures.

The first would be to run Hi/Lo and Mentor, each playing all the hands according to strict Basic Strategy. Since both systems have a 97% BC, their yields should be virtually the same, as long as the larger bets are all tripped in at nearly the same advantage points.

The second would be to divide all the Mentor card tags in half, so that the deuces, 7's, 9's and Aces are all plus or minus a half point, while the 3's thru 6's and 10's are plus or minus one point (this does not change the accuracy of the system). Then -- run the two sims again, both using the "count-per-deck" TC method, and both systems using the exact same set of index numbers! Since Hi/Lo has a PE of 51% and Mentor's is 62%, I can't imagine Mentor not doing better (even though some of the indices should be slightly different due to the 7 and the 9).

 

[QFIT]

It makes no sense to change Mentor to count-per-deck. In fact that would invalidate the comparison. Simply using the same indexes is close enough. And as I said, with the canned Sweet-16 sims, Mentor clearly outperforms level I strategies.

 

[Myth]

6D/S17/DSA/RSA with 1-12 spread and 75% Penetration all ran at 500 million rounds with CVCX

Zen (sweet16/fab4)
Score = 26.11
NO = 38,295
Win/HR = $24.22

Mentor (sweet16/fab4)
Score = 25.69
NO = 38,924
Win/HR = $24.48

RPC (sweet16/fab4)
Score = 26.16
NO = 38,233
Win/HR = $25.41

Hi-Lo (sweet16/fab4)
Score = 23.69
NO = 42,204
Win/HR = $23.01

 

[Renzey]

It makes no sense to change Mentor to count-per-deck. In fact that would invalidate the comparison. Simply using the same indexes is close enough. And as I said, with the canned Sweet-16 sims, Mentor clearly outperforms level I strategies.

Norm, I'm not sure we're understanding each other. Wouldn't the elimination of all the calibrating differences between Hi/Lo and Mentor (card tag scale, true count scale and index range) be the most homogeneous way to accurately compare the two? As another example, wouldn't it be clearest when comparing Zen with Mentor to convert Mentor to "count-per-deck", divide all its indices exactly in half (or do the inverse with Zen) -- then run the sims with optimum bets?

 

[Myth]

In those above sims I ran Mentor using the 2 Deck True Count, using the table provided in the Mentor book and CVCX's custom feature. All the rest used one deck.

 

[bjcount]

Norm, I'm not sure we're understanding each other. Wouldn't the elimination of all the calibrating differences between Hi/Lo and Mentor (card tag scale, true count scale and index range) be the most homogeneous way to accurately compare the two? As another example, wouldn't it be clearest when comparing Zen with Mentor to convert Mentor to "count-per-deck", divide all its indices exactly in half (or do the inverse with Zen) -- then run the sims with optimum bets?

Are you saying to convert one of the strategies so both are compared with the similar true count conversions, say both using 1dtc or both using 2dtc conversion? By doing this, then I would suggest doubling the indices of the strategy using 1DTC conversion (zen or hilo) because the slight benefit "granularity" offers to the 2DTC may skew the results in favor of the strategy originally generated with 1DTC.

BJC

 

[Renzey]

Are you saying to convert one of the strategies so both are compared with the similar true count conversions, say both using 1dtc or both using 2dtc conversion? BJC

Yes, as well as the card tag "scales" and index play TC limiters, so that nothing in the program allows anything to sneak thru the cracks due to those "out-of-the-box" differences.

 

[bjcount]

Are your index numbers for Mentor being "scaled" appropriately? Notice that its published indices scale them according to the "count-per-two-decks" TC method. Hence, the index for 15 vs. 10 is +15 TC. If you were to run the sim using the conventional "count-per-deck" TC, this play should kick in at +7 or +8 TC. Furthermore, if you're limiting the index plays that get tripped in by using a fixed TC number, even after making the appropriate TC adjustments, certain Mentor index plays will be will be deleted by mere virtue of its higher card tags.

There has to be something glaringly wrong with the Mentor sim results, and I'm not sure what it is. I can think of two cures.

The first would be to run Hi/Lo and Mentor, each playing all the hands according to strict Basic Strategy. Since both systems have a 97% BC, their yields should be virtually the same, as long as the larger bets are all tripped in at nearly the same advantage points.

The second would be to divide all the Mentor card tags in half, so that the deuces, 7's, 9's and Aces are all plus or minus a half point, while the 3's thru 6's and 10's are plus or minus one point (this does not change the accuracy of the system). Then -- run the two sims again, both using the "count-per-deck" TC method, and both systems using the exact same set of index numbers! Since Hi/Lo has a PE of 51% and Mentor's is 62%, I can't imagine Mentor not doing better (even though some of the indices should be slightly different due to the 7 and the 9).

After running about a dozen sims, I was going back over them to see if I missed something. One item I doubled checked was to see what how far apart the hands played were between HiLo and Mentor. As it turns out, per CVData sims I ran (not canned) using the I18 indices, a WO-2 point in HiLo played 42% of all hands. To achieve a similar WO point using Mentor I had to go down to TC-5 to play 48% of all hands. Does that sound about right?

BJC

 

[bjcount]

Yes, as well as the card tag "scales" and index play TC limiters, so that nothing in the program allows anything to sneak thru the cracks due to those "out-of-the-box" differences.

If you change the card tags then the indices will not be correct for the playing strategy. I tried changing the true count conversion on the playing strategy I use in real play to see if betting granularity made a difference. By switching from 1/2dtc to 1dtc the change was 0.50 cents increase/hr in the win rate. I didn't change any tags just the indices


BJC

 

[Myth]

I think all you need to do is compare by what their respective index numbers are set to. So for Zen 1 Deck TC and Mentor 2 Deck TC and I assume since your strategy is based on RPC it would be 1 deck also then simply run with same indexes so all ran using sweet 1f. Then you have an exact comparison, they all come out close to each other tho and beat Hi-Lo. So in the end its simply what ever sysem one would make less amount of mistakes with.

 

[Renzey]

After running about a dozen sims, I was going back over them to see if I missed something. One item I doubled checked was to see what how far apart the hands played were between HiLo and Mentor. As it turns out, per CVData sims I ran (not canned) using the I18 indices, a WO-2 point in HiLo played 42% of all hands. To achieve a similar WO point using Mentor I had to go down to TC-5 to play 48% of all hands. Does that sound about right?

BJC

I would've thought a Mentor -4TC would have you playing the same number of hands as a Hi/Lo -2TC -- if -- they were both using a 1dTC. I'll do some checking with my Wong B/J Count analyzer and report back.

 

[bjcount]

Just a quick note. I just checked the indices in BBII vs the packaged indices with CVData and there are a number of differences in the Soft DD and the Split tables. This may be affecting the results slightly, but I wouldn't believe the approx. 6-10 differences would make a great difference, but it is something to look at.

BJC

 

[QFIT]

Norm, I'm not sure we're understanding each other. Wouldn't the elimination of all the calibrating differences between Hi/Lo and Mentor (card tag scale, true count scale and index range) be the most homogeneous way to accurately compare the two? As another example, wouldn't it be clearest when comparing Zen with Mentor to convert Mentor to "count-per-deck", divide all its indices exactly in half (or do the inverse with Zen) -- then run the sims with optimum bets?

Fred, the only way to accurately compare two strategies is to compare them as they are in fact used. In fact, one of the problems with Zen is the manner in which it is described in its latest versions - and why Mentor is superior to the latest version of Zen. Oddly, most people that use Zen do NOT use it as it is described. And therefore it is inferior to what it could be. Peter Griffin was a genius. But his discussions were theory. I am interested in the details of usage.

Comparison of two strategies is a messy proposition. How do you compare two things when they have different methodologies? A prime example is your method for calculating true counts. I happen to like it. And, CVData is the only index generator that I know of that can actually handle it. Because I'm a bit nutty about how people actually use systems as opposed to theoretic advantages. (Comes from 16 years of customers telling me how they actually count.)

In any case, what I am saying is that tricks like trying to scale indexes and counting in some manner invalidate the comparison. What matters is what people actually do in a casino. And yes, Mentor clearly beats HiLo.

 

[zengrifter]

And yes, Mentor clearly beats HiLo.

Well, we all know that (except BJ Count, of course), but we are having trouble proving it in this thread. zg