#save pitcher data for simulation as csv #NOTE: the only csvs you need to download are clean_pitcherdata.csv and pitchnames_and_estsd.csv #(other spreadsheets were other versions no longer being used to run python/flask) library(tidyr) library(tidyverse) library(dplyr) #read raw data from 24 and 25 seasons s24 <- read.csv('statcast2024.csv') s25 <- read.csv('statcast2025.csv') #merge into one dataset all_data <- rbind(s24,s25) #find all pitchers that have enough data pitcher_options <- all_data |> group_by(pitcher, stand) |> summarize(observations = n()) |> filter(all(observations > 1000)) |> ungroup() |> distinct(pitcher) view(pitcher_options) #create clean data set for simulation pitcher_data <- all_data |> filter(pitcher %in% pitcher_options$pitcher) |> #only use data from pitchers that users can select select(pitcher, player_name, plate_x, plate_z, sz_top, sz_bot, #get necessary columns for simulation stand, p_throws, pitch_type, description, balls, strikes, pitch_name) #create lists #swinging strike qualifies as any outcome in this list swinging_strike <- c("swinging_strike_blocked", "swinging_strike", "missed_bunt", "automatic_strike") #ball is any outcome in this list ball <- c("blocked_ball", "ball", "hit_by_pitch", "automatic_ball", "pitchout") #foul is anything in this list foul <- c("foul", "foul_bunt", "foul_tip", "bunt_foul_tip", "foul_pitchout") #creating lists of pitch_type_groups #these groupings were listed on statcast fastball <- c("FF", "SI", "FC") offspeed <- c("CH", "FS", "FO", "SC") curveball <- c("CU", "KC", "CS") slider <- c("SL", "ST", "SV") knuckleball <- "KN" #creating the full clean dataset #already has data for only valid pitchers pitcher_data <- pitcher_data |> #create a column assigning pitch type to a pitch_type_group as listed above mutate(pitch_type_groups = ifelse(pitch_type %in% fastball, "fastball", ifelse(pitch_type %in% offspeed, "offspeed", ifelse(pitch_type %in% curveball, "curveball", ifelse(pitch_type %in% slider, "slider", ifelse(pitch_type %in% knuckleball, "knuckleball", NA))))), #using result lists from above - create possible outcomes simple_result = ifelse(description %in% swinging_strike, "swinging_strike", ifelse(description %in% ball, "ball", ifelse(description %in% foul, "foul", ifelse(description == "called_strike", "called_strike", description)))), is_swing = ifelse(simple_result %in% c("swinging_strike", "foul", "hit_into_play"), 1, 0), #did the batter swing yes (1) or no(0) is_contact = ifelse(simple_result %in% c("foul", "hit_into_play"), 1, 0), #did the batter make contact on the swing, yes (1) or no(0) is_foul = ifelse(simple_result == "foul", 1, 0), #was the contact a foul, yes(1) or no(0) is_swingstrike = ifelse(simple_result == "swinging_strike", 1, 0), #did batter swing and miss, yes(1) or no(0) is_hit = ifelse(simple_result == "hit_into_play", 1, 0), #did batter get a hit, yes(1) or no(0) #plate_x = ifelse(stand == "L", -plate_x, plate_x), #if the batter is a lefty, reflect the x-coord to be in terms of righty batters dist_x = plate_x, #dist from center horizontally dist_z = plate_z - 2.5) |> #dist from center vertically select(pitcher, player_name, plate_x, plate_z, sz_top, sz_bot, #select only necessary columns stand, p_throws, pitch_type, description, balls, strikes, pitch_type_groups, simple_result, is_swing, is_contact, is_foul, is_swingstrike, is_hit, pitch_name, dist_x, dist_z) |> drop_na() #save data set to csv write.csv(pitcher_data, "clean_pitcherdatatest.csv", row.names = FALSE) ############################################## #use pitch_options from above to view the valid pitchers and batters (thrown 1000 pitches) #create and save a csv of all pitch types by pitcher id p_options <- pitcher_data |> group_by(pitcher, stand, pitch_type_groups) |> select(pitcher, stand, pitch_type_groups) |> summarise(observations = n()) |> ungroup() final <- p_options |> filter(observations > 50) |> select(-observations) #using pitch groupings (fastball, offspeed, curveball) write.csv(final, "basicpitch_arsenal.csv", row.names = FALSE) #this may be a more appealing option and gives user more specific options to choose from #create and save a csv of all pitch types by pitcher id adv_p_options <- pitcher_data |> group_by(pitcher, stand, pitch_type_groups, pitch_type) |> select(pitcher, stand, pitch_type_groups, pitch_type) |> summarise(observations = n()) |> ungroup() #only keep pitch types that have been thrown at least 50 times for that pitcher, batter combination adv_final <- adv_p_options |> filter(observations > 50) |> select(-observations) #abbreviation of specific pitch thrown for each group write.csv(adv_final, "advpitch_arsenal.csv", row.names = FALSE) ########################### #FIXING PROBLEMS! #need to account for the count as well as type of pitch #need 50 pitches for each pitch type to run #need to make sure there is data for every type of count within these min 50 pitches #code that made current dataset... p_options <- pitcher_data |> group_by(pitcher, stand, pitch_type_groups) |> select(pitcher, stand, pitch_type_groups) |> summarise(observations = n()) |> ungroup() final <- p_options |> filter(observations > 50) |> select(-observations) #include count new_p_options <- pitcher_data |> group_by(pitcher, stand, pitch_type_groups, balls, strikes) |> select(pitcher, stand, pitch_type_groups, balls, strikes) |> summarise(observations = n()) |> ungroup() test <- new_p_options |> filter(pitcher == 542881, stand == "L", pitch_type_groups == "slider") #not about the count apparently problem <- pitcher_data |> filter(pitcher == 542881, stand == "L", pitch_type_groups == "slider") |> select(-c(plate_x, plate_z, sz_top, sz_bot, p_throws, pitch_type_groups, pitch_type, dist_x, dist_z)) |> filter(balls == 2, strikes == 0) #try this again knowing i need all counts and some variability test_data <- pitcher_data |> filter(pitcher %in% final$pitcher) |> #only get pitchers that are in final dataset select(pitcher, stand, pitch_type_groups, balls, strikes #, is_swing, is_contact, is_foul, #is_swingstrike, is_hit ) |> group_by(pitcher, stand, pitch_type_groups, balls, strikes) |> summarise(ob = n()) |> ungroup() #12 combinations of count... #only keep pitcher, stand, pitch combinations with 12 rows test_data |> group_by(pitcher, stand, pitch_type_groups, balls, strikes) |> filter(row_number() == 12, ob >= 50) test_final <- test_data |> distinct(pitcher, stand, pitch_type_groups) #try this again clean_test <- pitcher_data |> filter(pitcher %in% final$pitcher) |> group_by(pitcher, stand, pitch_type_groups, balls, strikes) |> summarize( total_pitches = n(), swings = sum(is_swing, na.rm = TRUE), no_swing = total_pitches - swings, foul = sum(is_foul, na.rm = TRUE), miss = sum(is_swingstrike, na.rm = TRUE), hit = sum(is_hit, na.rm = TRUE) ) |> filter(total_pitches >= 10, swings > 0 & no_swing > 0, ) |> group_by(pitcher, stand, pitch_type_groups) |> mutate(count_options = n()) |> filter(count_options == 12) |> ungroup() pitchers_clean_test <- clean_test |> distinct(pitcher, stand, pitch_type_groups) #only gives me fastballs... ##################### #getting sd for each pitcher, batter, pitch type combination variety_test_sd <- pitcher_data |> group_by(pitcher, stand, pitch_name) |> summarise( total_pitches = n(), swings = sum(is_swing), takes = total_pitches - swings, foul = sum(is_foul, na.rm = TRUE), miss = sum(is_swingstrike, na.rm = TRUE), hit = sum(is_hit, na.rm = TRUE), sd_x = sd(plate_x, na.rm = TRUE), sd_z = sd(plate_z, na.rm = TRUE), .groups = "drop" ) |> #additional filtering to make sure there is enough data for every situation filter(total_pitches >= 50, swings >= 25, takes >= 10, foul >= 10, miss >= 10, hit >= 10, sd_x > 0.2, sd_z > 0.2) #possible pitcher, batter, pitch matchups for user to choose from final_variety <- variety_test_sd |> distinct(pitcher, stand, pitch_name) #use original sheet dataset, find sd for each combinations pitch_sd <- pitcher_data |> group_by(pitcher, stand, pitch_name) |> summarize(sd_x = sd(plate_x), sd_z = sd(plate_z)) #join the possible selections with the sd for each pitch_join <- final_variety |> left_join(pitch_sd, by = c("pitcher", "stand", "pitch_name")) |> mutate(sd_x = round(sd_x, 3), sd_z = round(sd_z, 3)) #write csv of possible combinations with associated sds write.csv(pitch_join, "pitchnames_and_sd.csv", row.names = FALSE) ############################################ #when taking the sd for all pitches, we don't know wehre the pitcher was actually trying to aim #instead, try to estimate how much the pitcher is missing his "targets" by #we can use pitches near the edges of the strike zone and assume that they intended to throw outside of the zone #with the edge of the ball touching the edge of the zone (enough to be called a strike in ABS challenge systems) #LOGIC #assume that pitchers intended to throw so the ball caught the edge of the strike zone just outside of the zone #baseball radius ~ 0.12 ft (0.11 to ensure ball is on the line) #use data from one ball width (~0.22) inside the zone - 1 ft outside of the strike zone for all edges #target locations are 0.11ft outside of the strike zone for all edges #find the difference between the actual pitch location and the target location #run this for all pitchers - general miss_x and miss_z to fall back on if there is not enough data for other pitchers #only keep pitches that we can infer a target location (inside 1 ball - 1ft outside zone) miss <- pitcher_data |> filter(plate_x >= 0.7083 - 0.11 - 0.11 | plate_x <= -0.7083 - 0.11 - 0.11 | plate_z >= 1.5 - 0.11 - 0.11 | plate_z <= 3.5 - 0.11 - 0.11) |> filter(abs(plate_x) < 1.5, plate_z > 0.5, plate_z < 4.5) #max amount they miss is a foot if they truly intended on that target #add columns to the miss dataset miss <- miss |> mutate(target_x = ifelse(plate_x > 0, 0.7083 + 0.11, -0.7083 + -0.11), #store whether target was location outside L/R of plate target_z = ifelse(plate_z > 2.5, 3.5 + 0.11, 1.5 - 0.11), #store whether target was location outside T/B of zone miss_x = plate_x - target_x, #horizontal mixx miss_z = plate_z - target_z, #vertical miss closer = ifelse(abs(miss_x) < abs(miss_z), "L/R", "T/B")) #whether the pitch landed closer to the sides or top/bot of zone #sd of #pitches closer to left and right of zone general_miss <- miss |> filter(closer == "L/R") sd(general_miss$miss_x) #horizontal miss #0.389 #pitches closer to top and bottom of zone general_miss <- miss |> filter(closer == "T/B") sd(general_miss$miss_z) #vertical miss #0.331 #now, for each pitcher, batter, pitch combination, find the estimated sd_x and sd_z #find horizontal miss for each combination test_x <- miss |> group_by(pitcher, stand, pitch_name) |> filter(closer == "L/R") |> summarize(est_sdx = sd(miss_x)) #find vertical miss for each combination test_z <- miss |> group_by(pitcher, stand, pitch_name) |> filter(closer == "T/B") |> summarize(est_sdz = sd(miss_z)) #pitch_join from above holds the valid pitcher, stand, pitch combinations that work in the simulation #this join did not add any rows, but added columns for the estimated miss sds pitch_join_est <- pitch_join |> left_join(test_x, by = c("pitcher", "stand", "pitch_name")) |> left_join(test_z, by = c("pitcher", "stand", "pitch_name")) #remove old sds from dataset pitch_join_est <- pitch_join_est |> select(-c(sd_x, sd_z)) |> mutate(est_sdx = round(est_sdx, 3), est_sdz = round(est_sdz, 3)) view(pitch_join_est) #write new csv to use instead of old sd (pitchnames_and_sd.csv) write.csv(pitch_join_est, "pitchnames_and_estsd.csv", row.names = FALSE)