Генерация случайных распределений

Функции распределений в R

# Pattern: r<dist>(n, param1, param2, ...)
# d<dist>(x, ...) -> density/probability
# p<dist>(q, ...) -> cumulative probability
# q<dist>(p, ...) -> quantile

# Example with normal distribution:
rnorm(5, mean = 0, sd = 1)    # 5 random draws
dnorm(0, mean = 0, sd = 1)    # density at x=0
pnorm(1.96, mean = 0, sd = 1) # P(X <= 1.96)
qnorm(0.975, mean = 0, sd = 1) # z-score for 97.5%
cat('Four functions: r, d, p, q for each distribution')

rnorm(): нормальное распределение

set.seed(42)
# Standard normal (mean=0, sd=1)
z_scores <- rnorm(1000)
mean(z_scores)  # ~0
sd(z_scores)    # ~1

# Custom normal (IQ scores: mean=100, sd=15)
set.seed(42)
iq_scores <- rnorm(100, mean = 100, sd = 15)
mean(iq_scores)  # ~100
sd(iq_scores)    # ~15
summary(iq_scores)

# About 68% within 1 sd
mean(abs(iq_scores - 100) < 15)  # ~0.68

runif(): равномерное распределение

set.seed(7)
# Default: uniform on [0, 1]
u <- runif(5)
print(u)  # values in (0, 1)

# Uniform on [a, b]
temps <- runif(100, min = -10, max = 40)  # temperatures
mean(temps)   # ~15 (midpoint)
range(temps)  # should be within [-10, 40]

# Discrete simulation using floor()
dice_rolls <- floor(runif(10, min = 1, max = 7))
print(dice_rolls)  # integers 1-6
table(floor(runif(600, 1, 7)))  # roughly equal counts

rbinom(): биномиальное распределение

set.seed(123)
# 10 coin flips (prob=0.5), repeated 5 times
coin_flips <- rbinom(5, size = 10, prob = 0.5)
print(coin_flips)  # number of heads each time

# Bernoulli: single flip (size=1)
flips <- rbinom(20, size = 1, prob = 0.5)
print(flips)  # 0s and 1s
mean(flips)   # ~0.5

# Biased coin (prob=0.7)
biased <- rbinom(1000, size = 1, prob = 0.7)
mean(biased)  # ~0.7

# Many trials: normal approximation kicks in
hundred_flips <- rbinom(1000, size = 100, prob = 0.5)
mean(hundred_flips)  # ~50

rpois(): распределение Пуассона

set.seed(5)
# Calls received per hour (average 3)
calls <- rpois(24, lambda = 3)  # 24 hours
print(calls)  # integers, mostly 1-6
mean(calls)   # ~3 (expectation = lambda)
var(calls)    # ~3 (variance = lambda for Poisson)

# Count the distribution
table(calls)

# Rare events (lambda=0.5)
rare_events <- rpois(100, lambda = 0.5)
table(rare_events)  # mostly 0s and 1s
mean(rare_events)   # ~0.5

rexp(): экспоненциальное распределение

set.seed(42)
# Time between customer arrivals (rate=2 per minute)
# Mean wait time = 1/2 = 0.5 minutes
arrival_times <- rexp(100, rate = 2)
mean(arrival_times)  # ~0.5
sd(arrival_times)    # ~0.5 (mean = sd for exponential)

# Simulate a queue
cumulative_arrivals <- cumsum(rexp(10, rate = 3))
print(round(cumulative_arrivals, 3))

# Exponential CDF: P(X <= x) = 1 - exp(-rate*x)
pexp(0.5, rate = 2)   # P(wait <= 0.5 min)
mean(arrival_times <= 0.5)  # empirical estimate

sample(): дискретная выборка

set.seed(10)
# Sample without replacement (like drawing cards)
cards <- 1:52
hand <- sample(cards, size = 5, replace = FALSE)
print(hand)  # 5 unique cards

# Sample with replacement (bootstrap)
x <- c(10, 20, 30, 40, 50)
bootstrap_sample <- sample(x, size = 5, replace = TRUE)
print(bootstrap_sample)  # may have repeats

# Simulate rolling two dice 1000 times
dice <- function() sum(sample(1:6, 2, replace = TRUE))
rolls <- replicate(1000, dice())
table(rolls) / 1000  # empirical probabilities

Взвешенная выборка

set.seed(99)
# Loaded die: 6 is twice as likely
faces <- 1:6
weights <- c(1, 1, 1, 1, 1, 2)  # relative
norm_weights <- weights / sum(weights)

rolls <- sample(faces, size = 1000, replace = TRUE,
                prob = norm_weights)
table(rolls) / 1000
# 6 appears ~2/7 ~= 0.286 of the time

# Categorical sampling
categories <- c('A', 'B', 'C')
probs <- c(0.5, 0.3, 0.2)
sample(categories, 10, replace = TRUE, prob = probs)

rgeom() и rnbinom(): распределения количества

set.seed(42)
# Geometric: flips until first head (prob=0.3)
# Number of FAILURES before first success
flips_until_head <- rgeom(10, prob = 0.3) + 1  # +1 for the success
print(flips_until_head)
mean(flips_until_head)  # ~1/0.3 = 3.33

# Negative binomial: overdispersed count data
# e.g., number of parasites per host
counts <- rnbinom(100, size = 2, prob = 0.4)
mean(counts)   # ~3 (theoretical: size*(1-p)/p)
var(counts)    # much larger than mean -> overdispersed

# Compare variance to Poisson with same mean
pois_counts <- rpois(100, lambda = mean(counts))
var(counts) / var(pois_counts)  # > 1

rt() и rf(): t- и F-распределения

set.seed(1)
# t distribution: heavier tails than normal
t_vals <- rt(1000, df = 5)
mean(t_vals)  # ~0
sd(t_vals)    # > 1 (inflated by heavy tails)

# Compare to normal
n_vals <- rnorm(1000)
# t has more extreme values (heavy tails)
sum(abs(t_vals) > 3)   # e.g., ~30
sum(abs(n_vals) > 3)   # e.g., ~3

# F distribution: used in ANOVA F-tests
f_vals <- rf(1000, df1 = 5, df2 = 20)
range(f_vals)   # always positive
mean(f_vals)    # ~df2/(df2-2) = 20/18 = 1.11

Визуализация случайных распределений

set.seed(42)
x <- rnorm(10000, mean = 5, sd = 2)

# Quick summary statistics
summary(x)
cat('Mean:', mean(x), '\n')
cat('SD:', sd(x), '\n')
cat('Skewness (should be ~0 for normal):',
    mean(((x - mean(x))/sd(x))^3), '\n')

# Empirical quantiles vs theoretical
quantile(x, c(0.025, 0.25, 0.5, 0.75, 0.975))
# Compare to theoretical:
qnorm(c(0.025, 0.25, 0.5, 0.75, 0.975),
      mean = 5, sd = 2)

Быстрая проверка

Итоги: случайные распределения

set.seed(42)
# Quick reference of common distributions:
rnorm(3, mean = 0, sd = 1)        # Normal
runif(3, min = 0, max = 1)        # Uniform
rbinom(3, size = 10, prob = 0.5)  # Binomial
rpois(3, lambda = 3)              # Poisson
rexp(3, rate = 1)                 # Exponential
rt(3, df = 10)                    # Student t
rf(3, df1 = 5, df2 = 20)          # F distribution
rgeom(3, prob = 0.3)              # Geometric
sample(1:10, 3, replace = TRUE)   # Discrete uniform
cat('R has 20+ built-in distributions')