library(dendroTools) library(testthat) library(dplyr) data(LJ_daily_temperatures) data(example_proxies_1) MVA_parameter <- select(example_proxies_1, MVA) TRW_parameter <- select(example_proxies_1, TRW) # daily_response function should return a list with matrix and two characters test3 <- daily_response(response = example_proxies_1, env_data = LJ_daily_temperatures, method = "lm", metric = "adj.r.squared", lower = 250, upper = 251, previous_year = FALSE, row_names_subset = TRUE) expect_is(test3, "dmrs") expect_is(test3[[1]], "matrix") expect_is(test3[[2]], "character") expect_is(test3[[2]], "character") # stop functions were included to prevent wrong results expect_error(daily_response(response = TRW_parameter, env_data = LJ_daily_temperatures, lower = 200, upper = 270, fixed_width = -368)) expect_error(daily_response(response = example_proxies_1, env_data = LJ_daily_temperatures, method = "cor", lower = 250, upper = 270, previous_year = FALSE)) expect_error(daily_response(response = example_proxies_1, env_data = LJ_daily_temperatures, method = "cor", lower = 280, upper = 270, previous_year = FALSE)) # If row.names of env_data and response do not match, error should be given example_proxies_1_temporal <- example_proxies_1 row.names(example_proxies_1_temporal)[1] <- "1520" # random year is assigned # to row.name of the firest row expect_error(daily_response(response = example_proxies_1_temporal, env_data = LJ_daily_temperatures, method = "lm", lower = 260, upper = 270, previous_year = FALSE, remove_insignificant = FALSE)) # The order of data is unimportant, if row_names_subset is set to TRUE and # row.names are years. In this case, data is merged based on matching # row names. # will be ordered data data(example_proxies_1) MVA_parameter <- dplyr::select(example_proxies_1, MVA)