Predicting probability of hit from three variables

threevarwork.R

# see blog post at 
# https://baseballwithr.wordpress.com/2018/08/21/r-code-for-probability-of-hit-given-three-variables/

# load in tidyverse package

library(tidyverse)

# read in the 2017 statcast data

sc2017 <- read_csv("../StatcastData/statcast2017.csv")

# only look at balls in play and define the hit variable

sc2017 %>% filter(type == "X") %>%
  mutate(hit = ifelse(events %in% 
                        c("single", "double", "triple", "home_run"), 
                      1, 0)) ->  
  sc2017_ip

# define spray angle (correct Petit's reexpression)

sc2017_ip$spray_angle <- with(sc2017_ip, round(
  (atan(
    (hc_x-125.42)/(198.27-hc_y)
  )*180/pi)
  ,1)
)

# new def of spray angle that adjusts for side of batter

sc2017_ip$phi1 <- with(sc2017_ip,                             
                       ifelse(stand == "L",
                              -spray_angle, spray_angle))

# fit gam with three variables

library(mgcv)
fit <- gam(hit ~ s(launch_speed, launch_angle, phi1),
           data = sc2017_ip, family = binomial)
save(fit, file="threevarfit.Rdata")

predict(fit, data.frame(launch_speed = 90,
                        launch_angle = 10,
                        phi1 = 0))

invlogit(1.550209)
################################################

# here is the plotting part

library(tidyverse)
TH <- theme(plot.title = element_text(
  colour = "black", 
  size = 14, 
  hjust = 0.5, vjust = 0.8, angle = 0))

# set up a grid values of phi, ls, and la

phi_v <- seq(-45, 45, length.out = 100)
ls_v <- seq(80, 100, by=5)
la_v <- seq(-10, 30, by=5)
df <- expand.grid(phi1 = phi_v,
                  launch_speed = ls_v,
                  launch_angle = la_v)
invlogit <- function(x){exp(x) / (1 + exp(x))}

# find the predicted probability and define descriptive labels
# for the values of launch speed and launch angle

df$Probability <-  invlogit(predict(fit, df))
df$Launch_Speed <- factor(df$launch_speed,
                          levels = ls_v,
                          labels = paste(ls_v, "mph"))
df$Launch_Angle <- factor(df$launch_angle,
                          levels = la_v,
                          labels = paste("Launch Angle =",
                                         la_v))

# here is the graph

ggplot(df, aes(phi1, Probability, 
               color=Launch_Speed,
               group=Launch_Speed)) +
  geom_line() + facet_wrap(~ Launch_Angle,
                           labeller = label_value) +
  xlab("Adjusted Spray Angle (degrees)") +
  ylim(0, 1) + 
  scale_colour_brewer(palette = "Reds") +
  theme_dark() + TH +
  ggtitle("Fitted Probability of Hit as Function
  of Spray Angle, Launch Speed, and Launch Angle")