label statcast data as likely to be imputed

[bdilday]

Statcast releases data for which the values of launch angle and launch speed have been imputed, but do not include a flag to indicate which ones. This code uses a set of heuristics (essentially group by launch angle, launch speed, and event and label any that occur more than X times as likely imputed) to estimate which values are the imputed ones and provides a function to append a new imputed column to a data frame. The imputed values are read from a CSV file. A default based on my analysis is included in inst/extdata, but it is extensible if the estimates change or if the user has their own estimates.

[BillPetti]

Hey Ben, this is really cool.

Can you give more detail on how you determined which combinations of la, lv, event, and bb_type likely contain imputed values? I was also unclear why some of your default values for la and lv are 6 digits.

Also, you are importing an inner_join, but in the function you are using left_join. The parameter for the data also is different in the YAML compared to the function.

[bdilday]

Can you give more detail on how you determined which combinations of
la, lv, event, and bb_type likely contain imputed values?

Sure - see this script that shows how I derived the likely-imputed data,
https://gist.github.com/bdilday/c6628735d8ae6e0c1e85bfee78b0e7cc

this is a good reference also - I largely implemented the ideas here.
https://www.fangraphs.com/tht/43416-2/

I was also unclear why some of your default values for la and lv are 6 digits.

The launch angle and launch speed are multiplied by 10000 (or whatever you use for the inverse_precision variable) and rounded to integer. In other words I'm binning the data, in really narrow bins. The idea behind this was there'll be no ambiguity about whether floating point numbers are equal when we do counting and joining.

Also, you are importing an inner_join, but in the function you are using left_join.
The parameter for the data also is different in the YAML compared to the function.

Fixed.

[BillPetti]

Awesome, thanks!