{rstan} で 「状態空間時系列分析入門」 を再現したい。やっていること / 数式はテキストを参照。
リポジトリ: https://github.com/sinhrks/stan-statespace
library(devtools)
# devtools::install_github('hoxo-m/pforeach')
# devtools::install_github('sinhrks/ggfortify')
library(rstan)
library(pforeach)
library(ggplot2)
ggplot2::theme_set(theme_bw(base_family="HiraKakuProN-W3"))
library(ggfortify)
# モデルが収束しているか確認
is.converged <- function(stanfit) {
summarized <- summary(stanfit)
all(summarized$summary[, 'Rhat'] < 1.1)
}
# 値がだいたい近いか確認
is.almost.fitted <- function(result, expected, tolerance = 0.001) {
if (abs(result - expected) > tolerance) {
print(paste('Result is ', result))
return(FALSE)
} else {
return(TRUE)
}
}
ukdrivers <- read.table('../data/UKdriversKSI.txt', skip = 1)
ukdrivers <- ts(ukdrivers[[1]], start = c(1969, 1), frequency = 12)
ukdrivers <- log(ukdrivers)
ukseats <- c(rep(0, (1982 - 1968) * 12 + 1), rep(1, (1984 - 1982) * 12 - 1))
ukseats <- ts(ukseats, start = start(ukdrivers), frequency = frequency(ukdrivers))
確定的レベルと干渉変数のあるローカルレベルモデル。
model_file <- '../models/fig06_01.stan'
cat(paste(readLines(model_file)), sep = '\n')
data {
int<lower=1> n;
vector[n] y;
vector[n] w;
}
parameters {
# 確定的レベル
real mu;
# 確定的係数
real lambda;
# 観測撹乱項
real<lower=0> sigma_irreg;
}
transformed parameters {
vector[n] yhat;
for(t in 1:n) {
yhat[t] <- mu + lambda * w[t];
}
}
model {
# 式 6.2
for(t in 1:n)
y[t] ~ normal(yhat[t], sigma_irreg);
}
y <- ukdrivers
w <- ukseats
standata <- within(list(), {
y <- as.vector(y)
w <- as.vector(w)
n <- length(y)
})
stan_fit <- stan(file = model_file, chains = 0)
##
## TRANSLATING MODEL 'fig06_01' FROM Stan CODE TO C++ CODE NOW.
## COMPILING THE C++ CODE FOR MODEL 'fig06_01' NOW.
fit <- pforeach(i = 1:4, .final = sflist2stanfit)({
stan(fit = stan_fit, data = standata,
iter = 2000, chains = 1, seed = i)
})
stopifnot(is.converged(fit))
yhat <- get_posterior_mean(fit, par = 'yhat')[, 'mean-all chains']
mu <- get_posterior_mean(fit, par = 'mu')[, 'mean-all chains']
lambda <- get_posterior_mean(fit, par = 'lambda')[, 'mean-all chains']
sigma_irreg <- get_posterior_mean(fit, par = 'sigma_irreg')[, 'mean-all chains']
stopifnot(is.almost.fitted(mu, 7.4374))
stopifnot(is.almost.fitted(lambda, -0.26111))
stopifnot(is.almost.fitted(sigma_irreg^2, 0.0222426))
title <- 'Figure 6.1. Deterministic level and intervention variable.'
title <- '図 6.1 確定的レベルと干渉変数'
# 原系列
p <- autoplot(y)
# stan
yhat <- ts(yhat, start = start(y), frequency = frequency(y))
p <- autoplot(yhat, p = p, ts.colour = 'blue')
p + ggtitle(title)
title <- paste('Figure 6.2. Conventional classical regression representation of',
'deterministic level and intervention variable.', sep = '\n')
title <- '図 6.2 確定的レベルと干渉変数の古典的な回帰表現'
df = data.frame(drivers = as.numeric(ukdrivers),
seats = as.numeric(ukseats))
p <- ggplot(df, aes(x = seats, y = drivers)) +
geom_point() +
stat_smooth(method = 'lm', se = FALSE)
p + ggtitle(title)
title <- paste('Figure 6.3. Irregular component for deterministic level model',
'with intervention variable.', sep = '\n')
title <- paste('図 6.3 確定的レベル・モデルに',
'干渉変数がある場合の不規則要素', sep = '\n')
# テキストのタイトルは誤植
autoplot(y - yhat, ts.linetype = 'dashed') + ggtitle(title)