Example1_TSTS
Example1_FTS
afts1 <- createAFTS(tsts = example1_TSTS, fts = example1_FTS, na = FALSE)
plotPRD(afts1)
plotFAN(tsts = example1_TSTS, fts = example1_FTS, id = “Y1”, method = “A”)
plotFAN(tsts = example1_TSTS, fts = example1_FTS, id = “Y1”, method = c(“A”, “B”,…))
MAE, MdAE, MAPE,…
MASE, …
You can install the development version of the forvision package (containing the latest improvements and bug fixes) from github:
The TSTS schema:
library(forvision)
# После того как загрузили forvision package, встроенные данные в пакете сразу доступны
# В пакете пока 2 встроенных набора данных: example1_TSTS and example1_FTS
head(example1_TSTS)
#> # A tibble: 6 x 3
#> series_id value timestamp
#> <chr> <dbl> <int>
#> 1 Y1 3104. 1984
#> 2 Y1 3360. 1985
#> 3 Y1 3808. 1986
#> 4 Y1 4388. 1987
#> 5 Y1 4937. 1988
#> 6 Y1 5380. 1989The FTS schema:
head(example1_FTS)
#> # A tibble: 6 x 8
#> series_id method timestamp origin_timestamp forecast horizon Lo Hi
#> <chr> <chr> <int> <int> <dbl> <int> <dbl> <dbl>
#> 1 Y1 A 1989 1988 5406. 1 5183. 5630.
#> 2 Y1 A 1990 1988 5876. 2 5653. 6099.
#> 3 Y1 A 1991 1988 6345. 3 6122. 6569.
#> 4 Y1 B 1989 1988 5474. 1 5251. 5697.
#> 5 Y1 B 1990 1988 6010. 2 5787. 6234.
#> 6 Y1 B 1991 1988 6547. 3 6324. 6770.To create AFTS with NA values (by default) (rows in tsts with no match in fts will have NA values in the afts):
library(forvision)
afts1 <- createAFTS(tsts = example1_TSTS, fts = example1_FTS)
head(afts1, 10)
#> # A tibble: 10 x 9
#> series_id value timestamp method origin_timestamp forecast horizon Lo
#> <chr> <dbl> <int> <chr> <int> <dbl> <int> <dbl>
#> 1 Y1 3104. 1984 <NA> NA NA NA NA
#> 2 Y1 3360. 1985 <NA> NA NA NA NA
#> 3 Y1 3808. 1986 <NA> NA NA NA NA
#> 4 Y1 4388. 1987 <NA> NA NA NA NA
#> 5 Y1 4937. 1988 <NA> NA NA NA NA
#> 6 Y1 5380. 1989 A 1988 5406. 1 5183.
#> 7 Y1 5380. 1989 B 1988 5474. 1 5251.
#> 8 Y1 5380. 1989 C 1988 5406. 1 5183.
#> 9 Y1 6159. 1990 A 1988 5876. 2 5653.
#> 10 Y1 6159. 1990 B 1988 6010. 2 5787.
#> # ... with 1 more variable: Hi <dbl>To create AFTS without NA vaules using the option na=FALSE:
library(forvision)
afts2 <- createAFTS(tsts = example1_TSTS, fts = example1_FTS, na = FALSE)
head(afts2, 10)
#> # A tibble: 10 x 9
#> series_id value timestamp method origin_timestamp forecast horizon Lo
#> <chr> <dbl> <int> <chr> <int> <dbl> <int> <dbl>
#> 1 Y1 5380. 1989 A 1988 5406. 1 5183.
#> 2 Y1 5380. 1989 B 1988 5474. 1 5251.
#> 3 Y1 5380. 1989 C 1988 5406. 1 5183.
#> 4 Y1 6159. 1990 A 1988 5876. 2 5653.
#> 5 Y1 6159. 1990 B 1988 6010. 2 5787.
#> 6 Y1 6159. 1990 C 1988 5876. 2 5653.
#> 7 Y1 6877. 1991 A 1988 6345. 3 6122.
#> 8 Y1 6877. 1991 B 1988 6547. 3 6324.
#> 9 Y1 6877. 1991 C 1988 6345. 3 6122.
#> 10 Y2 4793. 1989 A 1988 4143. 1 3920.
#> # ... with 1 more variable: Hi <dbl>Эта функция использует схему AFTS без NAs в качестве входа
# Load the forvision package
library(forvision)
# To create AFTS without NA values from TSTS and FTS
df <- createAFTS(tsts = example1_TSTS, fts = example1_FTS, na = FALSE)
# Plot a PRD
plotPRD(df)
using xlim and ylim:
using useLog = TRUE
Эта функция использует TSTS and FTS schemas в качестве input
Column timestamp обязательно в виде time-based object
library(forvision)
# Преобразование сolumn timestamp в time-based object
example1_TSTS$timestamp <- zoo::as.yearmon(example1_TSTS$timestamp)
example1_FTS$timestamp <- zoo::as.yearmon(example1_FTS$timestamp)
# create a graph with one method
plotFS(ts = example1_TSTS, fs = example1_FTS, id ="Y1", meth = "A")Эта функция использует TSTS and FTS schemas в качестве input
Column timestamp обязательно в виде time-based object
# Using time series from M3
library(Mcomp)
#> Loading required package: forecast
# create TSTS
TS <- M3[[2000]]
TSTS <- data.frame(series_id = TS$st, value = c(TS$x, TS$xx), timestamp = c(time(TS$x), time(TS$xx)))
# Create models
library(forecast)
ets.model <- ets(TS$x)
arima.model <- auto.arima(TS$x)
nn.model <- nnetar(TS$x)
tbat.model <- tbats(TS$x)
# Forecast
ets.forecast <- forecast(ets.model,TS$h)
arima.forecast <- forecast(arima.model,TS$h)
nn.forecast <- forecast(arima.model,TS$h)
tbat.forecast <- forecast(tbat.model,TS$h)
# Create FTS
FTS <- rbind(cbind(series_id = TS$st, as.data.frame(ets.forecast)[,1:3], method = "ETS", timestamp = c(time(TS$xx)), horizon = 1:TS$h),
cbind(series_id = TS$st, as.data.frame(arima.forecast)[,1:3], method = "ARIMA",timestamp = c(time(TS$xx)), horizon = 1:TS$h),
cbind(series_id = TS$st, as.data.frame(nn.forecast)[,1:3], method = "NN", timestamp = c(time(TS$xx)), horizon = 1:TS$h))
rownames(FTS) <- NULL
colnames(FTS) <- c("series_id", "forecast", "Lo", "Hi", "method", "timestamp", "horizon")head(TSTS)
#> series_id value timestamp
#> 1 M599 2440 1979.000
#> 2 M599 2175 1979.083
#> 3 M599 1800 1979.167
#> 4 M599 1795 1979.250
#> 5 M599 2075 1979.333
#> 6 M599 1985 1979.417
head(FTS)
#> series_id forecast Lo Hi method timestamp horizon
#> 1 M599 5054.931 4202.243 5907.620 ETS 1989.500 1
#> 2 M599 5054.931 3849.108 6260.755 ETS 1989.583 2
#> 3 M599 5054.931 3578.130 6531.733 ETS 1989.667 3
#> 4 M599 5054.931 3349.682 6760.181 ETS 1989.750 4
#> 5 M599 5054.931 3148.414 6961.449 ETS 1989.833 5
#> 6 M599 5054.931 2966.454 7143.409 ETS 1989.917 6# Преобразование сolumn timestamp в time-based object
TSTS$timestamp <- zoo::as.yearmon(TSTS$timestamp)
FTS$timestamp <- zoo::as.yearmon(FTS$timestamp)
# create a fanplot
library(forvision)
plotFAN(ts = TSTS, fs = FTS, id ="M599", meth = "ETS")# for plotPRD-create AFTS
data <- createAFTS(tsts = TSTS, fts = FTS, na = FALSE)
# Plot a PRD
plotPRD(data)