What I need to do instead is find where there was a ball acquired, a throw (ball in play), and any number of call bounce events, and then ball acquired by the first baseman

I have found several in my spot checking, especially in 1903_17_TeamNI_TeamA3

• play 3
• play 187

My inspiration is the famous recursive sudoku solver. Here is some pseudo code

solve_outs_sequence(df, seq):
     if no more cells, return True

     for this_play_outs in range(3):
        if valid_assignment
             fill in seq here

             if solve_outs_sequence(df, seq), return True

             # this means you messed up
             remove those things you tried to add in seq

    return False

It could be helpful to add some context to the play

Should come up with a way to say whether distances/angles are positive of negative

• like for batter_dist, if could be that they overran the bag!

Compute how far "offline" a throw ends up, so I can map this to the same perspective

do this by added to the end of the file path

• plot all of the "ball bounce off wall" (or whatever it is) events
• plot all of the home runs too

What I notice isn't there that might be nice to have:

• the at_bat field is not there for older games
• I don't think I have any outs or count
  • Not sure I can get counts
• Players speed at a given time
• Player orientation?
  • I think I could noisily infer this
• I don't think that "ball bounce" is in the older data -- how is this defined?
  • I could maybe fill this one in if z_coord is < 0.5 or something arbitrarily small?

e.g. this one from "1903_01_TeamNE_TeamA2"

A really common data issue is having a second (but new) row for a given play_per_game in game_info

There is no way to know that the play is an out from that alone

game_info_1903_19.loc[(game_info_1903_19["inning"] == 7) & (game_info_1903_19["top_bottom_inning"] == "Top")]

there should be an out at 209

Right now I have hardcoded the fields I need, and I shouldn't

Because I am computing throw angle already, I can come up with descriptive findings of which of these tend to be errors

• I think its elevation, not angle usually

Can also just use the is_valid_assignment thing for each one

maybe also add the "trust this half" filter on all of these!
bf_1902_24 is a good one though

Successful looks like

Still need to filter though, see #10 and #6

g_1902_24_events = bf_1902_24.game_obj.game_events_df.copy()

# I should use that function I wrote here!

g_1902_24_events.loc[(g_1902_24_events["event"] == "ball acquired") & (g_1902_24_events["player_position"] == 3) &\
                     (g_1902_24_events["prev_event"] != "pickoff throw") &\
                     g_1902_24_events["play_per_game"].isin(game_info_1902_24.loc[game_info_1902_24["this_play_outs"] >= 1, "play_per_game"].values)
                     ,
                     :
                    ]

Unsuccessful and caught looks like

g_1902_24_events = bf_1902_24.game_obj.game_events_df.copy()

# I should use that function I wrote here!

g_1902_24_events.loc[(g_1902_24_events["event"] == "ball acquired") & (g_1902_24_events["player_position"] == 3) &\
                     (g_1902_24_events["prev_event"] != "pickoff throw") &\
                     g_1902_24_events["play_per_game"].isin(game_info_1902_24.loc[game_info_1902_24["this_play_outs"] == 0, "play_per_game"].values)
                     ,
                     :
                    ]

I am not sure what Unsuccessful and not caught would look like??

• Maybe look at the angle of the throw?

I see 14-19, and I bet those are the umpires, which I think would be kind of interesting tbh, but probably out of scope for this project

It looks like there are some plays that are repeated multiple times, see below

Right now, there are some occurrences where not everything happens at either the same cadence, or the same frequency. I imagine this will make analysis hard because I will have to do some fuzzy joins or something.

make a modules that lines up the data, but taking one series as ground truth (probably player pos?) and "snapping" other event to the nearest time stamp? Or maybe line up the data left joins into a single table, so I don't have to do deal with these conditional and this weird strobe effect that comes from the catcher being on the different phase as everyone else.

I would love to get like an umpire or first base coach view of some of these plays to first -- especially that air mailed throw to first (play 203, from 1903_01)

Maybe in a set of new plotting functions?

1903_01_TeamNE_TeamA2 is a good example of all of this

esp play_id 124!

I think I should have some kind of module stores the previous states

I don't remember what I did, double check the next time I go under the hood

I think the problem here is that the second baseman in ided as being wayyy left of the bag, but he tries to turn two.

See if I should do something about this based on game events or info?

A pick by Jeff McNeill on Jun 14, 2023 on a double play from a 118 mph ball hit by Stanton
https://twitter.com/mlb/status/1669157081377652736

Another data issue in 1902_24_TeamMA_TeamA1

• there are sometimes no BR listed in the game_info, even though there is definitely a person on base
  • examples:
    • 383 had someone get on but they aren't recorded
    • 386 shows a person on third in the data, but they aren't there in game info