Javascript simulation with graphs

[Cacarulo]

Thank you everyone for the feedback. Have tried to eliminate all sources of bias and increase efficiency. Decks at the desired count were found by walking through the deck and taking them as they come. From 1,000,000 to 100,000,000 hands per action as needed to get the EV margin of error below 0.003, and the difference between the top two actions to a confidence interval of 0.99 (typically a few are too close to call), about 7-10,000,000,000 hands per count. 25-45 minutes per count using an i9-12900H and 82 threads. The graphs look pretty smooth. (As to the previous comment about the RNG, it is JavaScript’s Math.random() in general, and Fisher-Yates for shuffling the deck.)

Here are the i18 and f4 indices that I get, calculating the true count by rounding the decks remaining (first index is rounding to whole deck, second is half deck). They are all the same as or one less than the ones in BJA3. If the indices themselves are rounded (rather than the current practice of flooring), all but a few are the same as in the book.

newIndices0213.png

In case anyone is interested . . . .

Your indices are much better now. Just a few small discrepancies (+/- 0.5) but overall they are correct. I'll pass you a table for 4D,S17,DOA,DAS,SPA1,SPL3.

CC Tags (A-T):   -1   1   1   1   1   1   0   0   0  -1

+-----------+-------+-------+--------+--------+--------+
|   Play    |       |       |   CA   |   CA   |   SIM  |
+-----------+-------+-------+--------+--------+--------+
| Insurance | -     | -     |   2.9  |    3   |   2.9  |
| 15vT      | Hard  | Stand |   4.2  |    4   |   4.0  | 4.7
| 16vT      | Hard  | Stand |   0.1  |    0   |   0.0  | 0.4
| 16v9      | Hard  | Stand |   4.5  |    5   |   4.4  | 4.5
| 12v6      | Hard  | Stand |  -1.0  |   -1   |  -1.0  |-1.2
| 12v5      | Hard  | Stand |  -1.5  |   -1   |  -1.5  |-1.5
| 12v4      | Hard  | Stand |   0.1  |    0   |   0.2  | 0.0
| 12v3      | Hard  | Stand |   1.7  |    2   |   1.7  | 1.5
| 13v3      | Hard  | Stand |  -2.2  |   -2   |  -2.2  |-2.4
| 12v2      | Hard  | Stand |   3.4  |    3   |   3.5  | 3.4
| 13v2      | Hard  | Stand |  -0.7  |   -1   |  -0.8  |-0.9
| 10vA      | Hard  | DD    |   3.5  |    4   |   3.3  | 4.0
| 11vA      | Hard  | DD    |   0.8  |    1   |   0.7  | 1.5
| 10vT*     | Hard  | DD    |   3.8  |    4   |   3.7  | 3.7
| 9v7       | Hard  | DD    |   3.4  |    3   |   3.3  | 3.5
| 8v6       | Hard  | DD    |   1.9  |    2   |   1.9  |
| 8v5       | Hard  | DD    |   3.5  |    4   |   3.4  |
| 9v2       | Hard  | DD    |   1.0  |    1   |   0.9  | 1.0
| A8v6      | Soft  | DD    |   0.7  |    1   |   0.6  |
| A8v5      | Soft  | DD    |   1.2  |    1   |   1.2  |
| TTv6      | Split | -     |   4.7  |    5   |   4.6  | 4.8
| TTv5      | Split | -     |   5.0  |    5   |   5.0  | 5.5
| TTv4      | Split | -     |   6.6  |    7   |   6.5  |
| TTv6      | Split | Das   |   4.7  |    5   |   4.6  | 4.8
| TTv5      | Split | Das   |   5.0  |    5   |   5.0  | 5.5
| TTv4      | Split | Das   |   6.6  |    7   |   6.6  |
| 15vA      | LSurr | -     |   1.5  |    2   |   1.4  | 2.0
| 14vT      | LSurr | -     |   3.1  |    3   |   3.1  | 3.4
| 15vT      | LSurr | -     |  -0.3  |    0   |  -0.3  |-0.2
| 15v9      | LSurr | -     |   2.4  |    2   |   2.3  | 2.3
+-----------+-------+-------+--------+--------+--------+

Sincerely,
Cac

[BackCounter]

Thank you again, Cac, for the data; it was a big help calibrating the system.

Performance improvements allowed reducing the EV margin of error to 0.001 in about the same amount of run time, 1,000,000 to 10,000,000,000 hands per action as needed, with the difference between the top two actions to a confidence interval of 0.9999. About 36,000,000,000 hands in 107 minutes, 100 threads.

All but two of the indices (16v10 = 1, 9v7 = 3) are the same as the "modern" indices on page 213. The catch-22 indices are also the same.

I calculated the EV difference by summing win rate difference from basic strategy * hand frequency * count frequency * units bet for each count (rather than approximating an integral as in the book). The hand frequency changes with the count. Compared to the book, R2 correlation is 0.926.

insurance 117 115
16vT 53 49
15vT 37 18
T,Tv5 17 24
T,Tv6 17 27
10vT 16 8
12v3 13 17
12v2 11 7
11vA 10 25
9v2 9 8
10vA 7 7
9v7 7 5
16v9 7 1
13v2 7 6
12v4 6 9
12v5 6 4
12v6 4 4
13v3 4 2