Prediction of 2019 HR counts using random effects model

home_run_prediction_2.R

################################################
## NEW WORK - August 2, 2019
################################################

#########################################
# data for 1631 games through August 1
#########################################

library(tidyverse)
library(CalledStrike)
sc2019 <- read_csv("~/Dropbox/2016 WORK/BLOG Baseball R/OTHER/StatcastData/statcast2019.csv")

# collect the number of home runs for each game

sc2019 %>%
  group_by(game_pk) %>%
  summarize(HR = sum(events == "home_run",
                     na.rm = TRUE)) -> S

## use poissonness plot to show lack of fit of P model

## 1631 games; 2430 - 1631 = 799 games to go

n_games <- 162 * 30 / 2 - nrow(S)

# bin the data and computing the vertical variable
# for the Poissoness Plot

S %>%
  group_by(HR) %>%
  summarize(N = n()) %>%
  mutate(p = N / sum(N),
         PPy = log(p) + log(factorial(HR)))  -> SU

# Poissoness Plot

  ggplot(SU, aes(HR, PPy)) +
  geom_point(color = 'red', size = 3) +
  geom_smooth(method = "lm", se = FALSE) +
    increasefont() +
    ggtitle("Poissoness Plot") +
    centertitle()

## fit random effects model by MCMC Stan

library(brms)
fit <- brm(HR ~ (1 | game_pk),
           data = S,
           family = poisson)

# collect the MCMC draws

posterior_samples(fit) -> B

# B is matrix of 4000 observations and 1503 variables

# here's the simulated predictions

replicate(1000,
  B %>%
  sample_n(size = n_games) %>%
  mutate(LAM = exp(rnorm(n_games, b_Intercept,
                         sd_game_pk__Intercept)),
         Y = rpois(n_games, LAM)) %>%
  summarize(HR = sum(Y) + sum(S$HR)) %>%
    pull(HR)) -> predict_HR

# graph the 2019 predicted HR counts

ggplot(data.frame(HR = predict_HR + sum(S$HR)),
                  aes(HR)) +
  geom_histogram(bins = 20,
                 color = "white",
                 fill = "orange") +
  increasefont()  +
  ggtitle("2019 Home Run Predictions") +
  centertitle()

#########################################