BuzzerBeater Forums

ok so ive actually come up with a way to work out the ratio of the 4 offensive skill sets in determining whether or not a specific shot type will go in.

Note that this will only take into account a specific shot type AND it can only take into account the 'chances of it going in' part. This will not help determine the 'chances of the Game Engine selecting that player to make that shot' part. Make sense?
ok.

So its a 4unknown simultaneous equation. With the unknowns, being the factor in which each skill is used in determining how its used for a specific shot type.
eg: Driving layups!

JSw + JRx + DRy + ISz = FG% (for driving layups).

And what I can do is actually just have a rolling set of data from players and pump it through a spreadsheet.

SO for instance, if i had Player X, I know his stats, and then I can take the FG% from 4 different games and end up with 4 equations.
Player X has JS = 15, JR = 11, DR = 15 and IS = 4.
Then i get his FG% (of driving layups) from 4 games, which is 50, 25, 33, and 45

Then my 4 equations end up like so.
15w + 11x + 15y + 4z = 50
15w + 11x + 15y + 4z = 25
15w + 11x + 15y + 4z = 33
15w + 11x + 15y + 4z = 45

SO problem is, is that equation won't give me the correct numbers because of the other factors which are:
- Offense selected
- Defense select of the opponent
- OPpositions player stats? (eg: OD/ID/SB?)
- Players gameshape
- Stamina?
etc

BUT, the beauty about it, is that if i start building a catalogue of data, I can eliminate some of those extra factors and start to get cleaner results.
eg: Capture the Gameshape, capture the offense/defense, and then standardise the data - that is, only use Strong Gameshape and Base Offence vs M2M.

Then i could even show that relationships between the factors and selected offenses, which could potentially lead to showing that (for example) Base offence utilises more DR than IS on a driving layup.

I wonder how much data id need though, thats the question. This could be a season long data collection exercise.

Thoughts?

A recent suggestiin post around the play by play asked for several things, one being showing the ball move around the court as it gets passed around.

I proposed that the GE does not work like that and that it only uses max 3 players during an offensive 'set'.

Howdy. Fascinating thread you have going here, a tantalizing mixture of observation, quasi-statistics and pure speculation, with a dash of illogical provocation to spice it up. I love it.

I would like to ask a couple of general questions, going back to square one ...
1. Do you have any data or even observation to underlie your thought that only 3 players are used in an offensive possession? That is contrary to a couple of things BB-bigshots (I don't exactly know what you call them) have clearly stated.

A) Bring the bal up the court
B) after halft court plays.

2. Do you have any data or even observation to underlie your thought that back court offense and front court offense are strictly separated, or should we take this as either speculation or a way to compartmentalize the analysis?

3. Where is the conversation about the effects of O-tactic vs D-tactic in general and defensive matchups in particular?

Thanks!

Last edited by Mike Franks at 7/4/2014 1:01:43 PM

everything ive stated is pretty much an assumption. but im trying to use fairly simple logic, specifically from a process control perspective and general maths.

So I split the two events initially, based on the fact that, before shots are taken, selected events happen, which are
- steal/turnover/halfcourt prayer/foul.

Now i say 'Bring the ball up the court' but in reality, thats describing 'all events before the shot selection engine'.

So for instance, certain events will occur before a shot selection engine is entered (in my opinion). like fouling at end of game as soon as possession is lost/rebound, timeouts etc etc....

It would be very inefficient for the game engine to run through an entire shot selection/passing/matchups etc etc, and then at the end, take into account the shot clock and exit with a timeout.

Its more likely that that Game engine takes in some initial data, and then enters a pathway.
eg: If game clock is greater than 2 mins, run normal process. If less than 2 mins, enter 'low-time process'.
.... --Normal Process
......... - Then if shot clock is greater than 20, run half court process
................. - Half court process
..........................runs some numbers for determining steals/timeouts/turnovers etc.
- if no event, then exit half court process and enter Shot selection process.
etc.
Not sure how that formatting is going to come out.

Thats purely my view, i have no stats or anything to back that up, its more just thinking about how you would write the whole thing.

With regards to the 3 players involved in the shot selection process, thats just my thoughts on this and can be discussed at leisure.

cool. thanks heaps for that summary. Ill take a few reads to digest that.

In the meantime, found these stats on the NBA.com website.

http://stats.nba.com/playerTrackingShooting.html?pageNo=1&rowsPerPage=25&SeasonType=Regular%20Season&filters=GP*GE*20**MIN*GE*15&sortField=PTS_DRIVE&sortOrder=DES

Almost exactly like BB!

I have a question.

What happens with player with JR>JS according to your theories?

Remove LI and GDP.

Why,
Then LP will be dominant or Patient.
If a team plays the same offense every game should be punished.