Postby solarkev » Thu Oct 27, 2022 3:33 pm
I've been working on new playbooks lately, and have been wondering about the computer play calling. So, I decided to make a couple test playbooks (attached) where each play is given one and only one PBAI tag, while having a uniquely numbered name for easy recognition. I used a test playbook with the "ask Madden" feature to determine how the computer AI chose plays. This method is similar to Ben75's work from this older thread, but using many plays per PBAI tag instead of just one, as in his case. Since this turned out to be more or less a confirmation and maybe a bit more elaboration of his findings, I thought it best to report here in his thread so that everyone interested could evaluate and compare.
Findings:
1. As Ben75 indicated, the descriptions in Packfan's editor are not completely accurate. However, this does little to diminish the brilliance of his playbook IMHO.
2. I didn't come across (yet) every tag that Ben75 found, but my results are nearly the same as he indicated for most of the tags.
3. The tags are used based on down & yards to 1st down, field position, and in a few cases, score and time left in the half.
4. It looks like the computer selects a PBAI tag for a certain situation and then randomly selects a play with that tag.
5. If play type is used at all, it is secondary to random selection. This seems proven by comparing the performance of the two attached playbooks. The major difference between these test books is that the 1st down and second down tags (PBAI 38 and PBAI 6, respectively) are populated by >75% passes in the first and >75% runs in the second. In the first case, passes were selected overwhelmingly, while in the second runs were selected overwhelmingly, both roughly in proportion to the ratio of pass plays versus running plays.
6. Here is how the computer used the tags in my tests (3 full games).
For 1st down with 10-14 yards to go for 1st down, from about the 10 yard line to the opponent’s 20, PBAI 38 is used >95% of the time. The few other instances used PBAI 6.
For 2nd down with 1-14 yards to go, from about the 10 yard line to the opponent’s 20, PBAI 6 was used all the time.
For 3rd down with 0-3 yards to go, except near the opponent’s goal line, PBAI 1 or 40 was used.
For 3rd down with 4-8 yards to go, PBAI 7 was used.
For 3rrd down with 9 yards or greater to go, PBAI 8 was used.
For 1st or 2nd down In the red zone (opponent’s 20 to the 5), with more than 3 yards to go, PBAI 4 was used.
Inside the opponent’s 5, PBAI 3 was used.
With less than 2 minutes in the 1st half with the lead, I saw PBAI 13 used.
Late in the half and trailing the score, I saw PBAI 10 used in the final 2 minutes. While trailing more than a touchdown, I saw this tag used on second and third down starting around 5 minutes or so in the second half.
I saw PBAI 21 used a couple times in the final minute of the half while trailing in the score.
The computer always punted or tried a field goal on 4th down and always tried to kick the extra point (never went for 2).
No other tags were ever used in these 3 test games.
Conclusion: If making offense playbook PBAI tags, I would use all of Ben75's definitions (I'm assuming that his findings about PBAI 2, 5, 34, and 37 are also valid, though I never saw them used).
Feel free to play around with the attached playbooks used for testing). I hope I explained things clearly, but there is a lot of info here. So, please pepper me with questions or comments, if you are so inclined.
-
Attachments
-
- LA Rams 2023 PBAI 2nd test.pbo
- (191.19 KiB) Downloaded 108 times
-
- LA Rams 2023 PBAI 1st test.pbo
- (191.19 KiB) Downloaded 110 times