tests/testthat/test-bsts.R

In davidtedfordholt/fable.bsts: Bayesian Structural Time Series Modelling Interface for 'fable'

context("test-bsts")
library(dplyr)

# test_that("bsts simple", {
#   default <- model(tsibble::as_tsibble(USAccDeaths), BSTS(value ~ season("year")))
#   expect_s3_class(default, "mdl_df")
#   default_mdl <- default[[1]][[1]]$fit$model
#   expect_length(default_mdl$seasonalities, 1)
#   expect_length(default_mdl$changepoints, 25)
#   expect_length(default_mdl$extra_regressors, 0)
#
#   default_fc <- forecast(default, h = 17)
#   expect_s3_class(default_fc, "fbl_ts")
#   expect_equal(NROW(default_fc), 17)
# })
#
# test_that("bsts complex", {
#   skip_if_not_installed("tsibbledata")
#   vic_elec <- tsibbledata::vic_elec %>%
#     filter(lubridate::year(Time) == 2014)
#   elec_tr <- vic_elec[1:(24*7*5),]
#   elec_ts <- vic_elec[(24*7*5 + 1):(24*7*7),]
#   aus_holidays <- tsibble::tsibble(
#     holiday = c("New Year's Day", "Australia Day", "Good Friday",
#                 "Easter Monday", "ANZAC Day", "Christmas Day", "Boxing Day"),
#     date = structure(c(16071, 16097, 16178, 16181, 16185, 16429, 16430), class = "Date"),
#     index = date)
#   complex <- model(elec_tr,
#                    fit = BSTS(Demand ~ trend("linear") +
#                               season("week", 3) +
#                               season("year", 12) +
#                               Temperature +
#                               holiday(aus_holidays))
#   )
#
#   expect_s3_class(complex, "mdl_df")
#   complex_mdl <- complex[["fit"]][[1]]$fit$model
#   expect_named(complex_mdl$seasonalities, c("week", "year"))
#   expect_length(complex_mdl$changepoints, 25)
#   expect_named(complex_mdl$extra_regressors, "Temperature")
#   expect_equal(complex_mdl$holidays$holiday, aus_holidays$holiday)
#
#   complex_fc <- forecast(complex, elec_ts)
#   expect_s3_class(complex_fc, "fbl_ts")
#   expect_equal(NROW(complex_fc), 24*7*2)
# })