BACK	Our   Roulette Timing Data

30 August 1997

The files we have made are of the format shown below, for the most part. One sample trial is shown here. The real datasets contain many trials; only one trial is shown in Table 1 below. The data were collected circa 1984. Trials are terminated by a checksum. The files, or datasets, contain Ball and Rotor timings, the winning pocket number [WON] in which the ball landed in the trial, the (elapsed) time at which the ball landed, the observed "would-have-landed" pocket [WLD] if there had been no "bounce" of the ball, and the predicted winning pocket, using our algorithm [PC2]. (Our algorithm was based on the assumption that the decay of the ball is a second-degree polynomial.
(The constants a, b and c, for 2^nd-degree polynomial y = ax^2 + bx + c were determined empirically prior to play – using files that contain only BALL data. These data gathered over a period of several hours prior to play and were analyzed using curve fitting routines. There routines determined the constants a, b and c for the polynomial).

Some of the files we have contain only Ball data -- ball revolution times with the winning pocket number. Some files contain the rotor revolution times and additional items.

The BALL and ROTOR (We also use the term "WHEEL" for ROTOR) were timed against an arbitrary spot on the outside of the rim of the roulette wheel (12:00 O’clock).

Datasets of the format shown below contain timings of the ball and the wheel; similarly formatted datasets contain only timing of the ball. (Ball timings were deemed most important to analyze – we discovered that ROTOR timings, even over very long times, were fairly steady (the decay of the wheel was very regular and almost linear).

Numbers in array#1 and array#2 represent time for one revolution, in milliseconds. The timing of the ROTOR (array#2) started the clock running – all other times are relative to this start time.

Although the BALL times appear first in the array, they are captured AFTER gathering the ROTOR times.

After the croupier launched the ball, the ROTOR times were gathered first. This allowed the ball to dissipate some energy. In most cases, the staring of timing the ROTOR is indicated by the "1" in the ROTOR data (array#1). In some cases, if many ROTOR times were obtained, only the latest times were saved in array#1. The array elements were pushed up, and the earliest clock times were pushed off the top of the array; thus "1" may not always appear as the first entry in array#1.

The same "push-up" principle is in effect for array#2, (the timing of the BALL); popping old timings off the top of the array preserves the latest timings and discards the oldest ones.

For BALL and ROTOR clock times, subtract successive clock-times to obtain revolution time (in milliseconds). The last entry in the Ball array usually represents only a partial revolution of the BALL, since this is the time when the ball entered a pocket (or was deflected by a pocket partition).

In the case of deflections by pocket partitions or "diamonds", we believe our data also shows the pocket in which the ball would have landed if no deflection had occurred. Our data are 13 years old, at time of this writing, and this is, to the best of our recollection, what data element #17 in the arrays means.

Data were gathered using American wheels having 00 and 0 pockets. All trials were conducted on roulette wheels in which the ball was spun in the clockwise direction, and the rotor in the opposite direction. Atmospheric pressure and ball diameter and condition were not noted. The sex if the croupier was not recorded.

Our data files probably cannot be combined for analysis. They represent individual sessions with specific wheels. We do not have information that shows which data files were collected on the same day, with the same ball and wheel, the same croupier, etc.

The data element [WON] may appear in a converted form on some files. A conversion table is shown here. The for each wheel pocket number in the first column, the corresponding clock position is shown the third column. The sequential order of the pockets is shown in column two. Converted data is recognizable by the two-decimal figures. For example, 6.95 represents a clock-position on the rim of the roulette wheel where the ball landed (roughly at seven o’clock; this represents a winning pocket numbered 9).

These data are available for your analysis. Please let us know something about your project and use of the data.

We are interested in any insights recipients of these data can provide. We are interested in neural net analyses of these data. We thank you kindly in advance for your input and cooperation.

Pictures from our roulette prediction project

CONVERSION TABLE

Table 2

SAMPLE DATA

---------------First Trial starts here---------------------

0 > Time when the ball passed the same fixed location going in
0 > the other direction. This array also bubbles up to take in
6122 > the new times. The last location (array 8) contains the
7116 > 'touchdown time' - the time the ball landed in the wheel.
8260 >
9569 >
11116 >
12665 > ("touchdown time" of ball - time the ball landed)
0 << these six numbers are the times when the "00" location on
0 << the wheel passed a fixed point. the clock always starts
1 << with '1', which is captured at the first press for the new
1590 << trial at time "0." As new times are entered, the array is
3163 << rolled up one location to a maximum of 6 for the wheel.
4782 << Times are in milliseconds.
4 -- The winning number.
10 xx The clock position in which the ball landed.
8 == Unknown - observed would have landed if no bounce???
32 @@ Predicted winner using our algorithm.
64438 \\ checksum - sum of above 18 numbers
---------------Second Trial starts here---------------------
0
0
0
7024
8242
9702
11437
12060
0
1
1465
2981
4506
6026
10
6
8
35
63503
0
0
5435
6580
7874
9398
11254
11856
0
0
1
1600
3273
4966
7
3
6
28
62281
0
0
0
6162
7404
8867
10676
11454
0
1
1493
2995
4484
5978
3
4
3
13
59537
0
0
0
5678
6942
8430
10208
11001
0
0
1
1752
3528
5289
14
4
14
33
52894
0
0
0
5703
6938
8377
10149
10725
0
0
1
1522
3057
4614
2
3
6
0
51097
0
0
0
5908
7068
8430
10006
11352
0
0
1
1710
3462
5182
4
7
10
29
53169
0
6642
7596
8699
9996
11514
13423
13789
0
1
1277
2551
3808
5128
4
1
3
30
84462
0
0
5057
6068
7201
8516
10040
11515
0
0
1
1364
2745
4126
10
9
10
2
56664