BuzzerBeater Forums

Do you by chance happen to have the defense these offenses faced?

Yes for all games I both have the play by play (where there is no information on the tactics) and the boxscore, which I download through the BBApi. In the boxscore I get all kinds of information on tactics, GDP, effort etc.

I need the boxscore data both to check that the totals of the play by play are correct and to allocate the steals from one team to the players of the other team as turnovers, because the play by play does not track who gets stripped.

However as you can see from the following pic, boxscores in BBApi have all the info on the game and all the traditional stats:
http://tinypic.com/r/2emieer/9

There is both delta effort and (TIE vs CT = -2, Normal vs TIE = +1 etc), defense used and GDP focus and pace (N/A = not used GDP).

Last edited by Lemonshine at 4/19/2018 4:54:40 AM

Making tables by position will take quite a lot of time. The overall table runs quite fast but when you add more and more criteria (like the tactic or the position who took the shot) it slows the code quite noticeably. I need to rework the code before I can do this, so that the dataset used for each query is as small as possible.

Note that the boxscore will not tell you where a player was playing when he took a shot (although most players only play at 1 position) while from the pbp I build the lineups for each action that happens (from starting lineup following with swaps and substitutions) so I could actually do this properly (i.e. if a player has 24 minutes at PG and and 24 at C I know which shots were taken at PG and which were taken at C).

I need to rewrite the code because the pbp dataset has 121k lines (most games have between 450-550 lines) and 56k are shots or FTs.

Last edited by Lemonshine at 4/18/2018 4:54:06 PM

May I ask what language you are using to scrape the BBAPI? I've been working on building something using Python to scrape the data from BBAPI but it's in the early stages of design, such as logging in, getting the basic Team Info and throwing it into a dictionary.

Before I go any further, I need to sit down and map out what I want to achieve with the data and then start planning the functions from that point forward.

Anyways, I don't want to hijack your thread and it's purpose so I'll leave that for a discussion later.

First of all thanks the author for this massive information, great work.
I have a question though. Anybody any idea how in this game PPP of my C is less than my opponents'??
http://digitalfieldtheory.com/match/101858254/boxscore.aspx
His centers went 1-5 FG 1-1 FT (Starting center 0 points in 36mins) total of 3 Pts and my Centers were 5-10FG 1-4 Ft - 11 points in total. Still, PPP for my C was 111 compared to my opponents 112,4. Very strange.

Last edited by Nicolas at 4/19/2018 1:50:30 AM

great job!

it seems base tactic is the one which provides the highet % of TS%. correct?

I'm not that advanced with coding, I do modelling in VBA for work so I used VBA (also this allowed me to debug all the problems which may happen infrequently, such as Lucky Fans getting into the action. I just used the same logic behind Moutlinho's tool which was coded in VB (I believe), but I never got that to work properly. I don't know how he solved the problem of the turnovers which seems impossible with just the pbp.

My tool simply does this: it takes the page source of the play by play and breaks down each entry. Instead of copying and pasting 1 at a time for 200 games, I get VBA code to do it for me. Using the page source rather than the text also makes it easy to identify the action as the same action can have different text in the pbp.

The BBApi is a much much faster and easier thing to download. After all this is what Buzzer-Manager also does. All the info is there in xml format and it does not need you to go to a navigate to webpage, all you need to do is parsing the information you want from the xml text.

Last edited by Lemonshine at 4/19/2018 10:27:24 AM

Yes but keep in mind a couple of things:
1) some tactics have a low sample as Teemo said
2) 'weaker' tactics like BO, R&G, outside iso etc are often played when the result is not in question, so when there is a disparity between teams and the stronger one plays an unusual tactic to prevent GDP

The number of possessions and which position is taking which kind of shot are also key to understand what exactly is going on. All the teams in the KO phase of the B3 may be more 'specialised' than your average BB team. So people playing outside tactics consistently may have built their team to be truly elite in some departments.

This analysis does not show you the baseline % for each tactic, it only shows the result of specific games. I suppose we can take some comfort that the numbers don't seem to penalise outside tactics so much, but again these numbers only cover one facet of the game, they will tell you nothing about rebounding, about turnovers, about pace and it is very likely that offensive tactics also have an impact in all of these other aspects. If we wanted to analyse the baseline it would be better to take a few D4 leagues with no bots than the cream of the crop because then you will have teams filled with a lot of average players.

Last edited by Lemonshine at 4/19/2018 10:25:28 AM

I chose this data since BBB KO games are high level games, played at 12 TS / same attitude since s34 and does not suffer from HCA influence. I think this is the best data available to analyze. Indeed GDP influence the output data, but this is also a present reality of BB : you have way more chances to be predicted inside.

This data should be compared to England first division data to be validated since the number of occurences of some tactics are low and uncertainties about them are probably high. I will try to find some interesting outside offense BBB round robin games to add to the database in order to reduce these uncertainties.

I just want to share an idea, not so sure how good/bad it is but I'm confident you and Lemonshine can :-)

do you think it could be useful taking a new set of data from games bewteen BOT teams only? It is true that there is the HCA influence, but you will have a huge set of data, and the low level of the players can give you minimal variance from the medium values.
I don't know if there are series made of BOT teams only. You can also use only data from matches with few points of difference in the final result. And these data can be a sort of "reference" for % values of interesting parameters.