Simulate Tossing Coin Game
Feb 24, 2019 #monte-carlo
The use of simulation experiments to better probability patterns is called the Monte Carlo method. This is an example.
Tossing Coin Game
Tom and Jerry play a simple game of tossing a coin. If given heads, Tom pays Jerry a dollar. Otherwise, Jerry pays Tom a dollar.
set.seed(1212)
# how many tosses?
number_of_tosses = 50
# simulate n tosses
# 1 = head, -1 = tail
(tosses <- sample(c(1, -1), size = number_of_tosses, replace = TRUE, prob = c(.5, .5)))
## [1] -1 -1 1 -1 1 -1 -1 -1 1 -1 1 1 -1 1 1 -1 -1 -1 1 -1 1 -1 -1 -1 1
## [26] -1 1 1 1 1 -1 -1 -1 1 -1 1 1 1 1 -1 -1 1 1 -1 -1 1 -1 1 1 -1
Let’s observe some games’ progression, from the perspective of Tom.
library(purrr)
# convert game to a function
new_game <- function(n = 1) {
tosses = sample(c(1, -1),
size = n,
replace = TRUE,
prob = c(.5, .5))
}
# progress of 4 different games
par(mfrow = c(2, 2))
walk(1:4, ~ {
plot(cumsum(new_game(number_of_tosses)), type = "l", ylab = "Winnings")
abline(h = 0, col = "red", lty = 3)
})
Let’s calculate the winnings distribution.
library(dplyr)
library(ggplot2)
library(ggthemes)
old <- theme_set(theme_tufte() + theme(text = element_text(family = "Menlo")))
# modify function above
new_game <- function(n = 1) {
tosses = sample(c(1, -1),
size = n,
replace = TRUE,
prob = c(.5, .5))
# return winnings
sum(tosses)
}
# repeat many times
reps <- 10e3 %>% rerun(new_game(number_of_tosses)) %>% unlist()
# bind into data frame
reps %>%
as_data_frame() %>%
ggplot(aes(value)) +
geom_histogram(binwidth = 1) +
scale_x_discrete(limits = seq(-24, 24, 4))
## Warning: `as_data_frame()` is deprecated as of tibble 2.0.0.
## Please use `as_tibble()` instead.
## The signature and semantics have changed, see `?as_tibble`.
## This warning is displayed once every 8 hours.
## Call `lifecycle::last_warnings()` to see where this warning was generated.
Question: Why there isn’t any odd number?
We can also calculate the chances where Tom break even, either from simulation result above, or using binomial distribution. They are pretty close.
# how many breakeven
length(which(reps == 0)) / length(reps)
## [1] 0.1163
# approx. form using binomial distribution
dbinom(25, size = 50, prob = 0.5)
## [1] 0.1122752
What is the likely number of tosses Peter will be winning in a game of 50 tosses? What is the likely maximum fortune Peter will be winning in a game of 50 tosses?
library(tidyr)
# modify function above
new_game <- function(n = 1) {
tosses = sample(c(1, -1),
size = n,
replace = TRUE,
prob = c(.5, .5))
# track progress
cum.win = cumsum(tosses)
# return
list(
final = sum(tosses),
leads = sum(cum.win > 0),
max = max(cum.win)
)
}
new_game(number_of_tosses)
## $final
## [1] 12
##
## $leads
## [1] 50
##
## $max
## [1] 13
# repeat
set.seed(1234)
reps <- 10e3 %>% rerun(new_game(number_of_tosses))
reps %>%
bind_rows() %>%
# collapse columns into one
gather(var, val) %>%
ggplot(aes(val)) +
geom_histogram(binwidth = 1) +
facet_wrap( ~ var, ncol = 1,
labeller = labeller(
var = c(final = "Final Winnings",
leads = "# Leadings",
max = "Max Fortunate")
)) +
labs(x = "", y = "")