rtime-seriestidymodelsr-recipestimetk

Error in step_fourier from timetk package

I have the following error when using the step_fourier function. This error only occurs when I use time_series_split split with my data to convert paneled time series into training and testing sets. When I input the dataset directly, the error does not occur.

Here the error occurs:





library(recipes)
library(tidyverse)
library(tidyquant)
library(timetk)


dados <- FANG  %>% select(symbol,date,volume)

splits <- dados %>% 
  time_series_split(
    assess     = "1 months", 
    cumulative = TRUE
  )



# Create a recipe object with a timeseries signature step
rec_obj <- recipe(volume ~ ., data = training(splits)) %>%
  
  step_fourier(date,period=365, K = 1)



rec_obj %>% prep() %>% juice() %>% glimpse()

Follow the error:

    Rows: 3,948
    Columns: 5
    $ symbol         <fct> FB, AMZN, NFLX, GOOG, FB, AMZN, NFLX, GOOG, FB, AMZN,…
    $ date           <date> 2013-01-02, 2013-01-02, 2013-01-02, 2013-01-02, 2013…
    $ volume         <dbl> 69846400, 3271000, 19431300, 5101500, 63140600, 27509…
    $ date_sin365_K1 <dbl> NaN, NaN, NaN, NaN, NaN, NaN, NaN, NaN, NaN, NaN, NaN…
    $ date_cos365_K1 <dbl> NaN, NaN, NaN, NaN, NaN, NaN, NaN, NaN, NaN, NaN, NaN…
    Warning messages:
    1: Problem while computing `date_sin365_K1 = timetk::fourier_vec(...)`.
    ℹ NaNs produced 
    2: Problem while computing `date_cos365_K1 = timetk::fourier_vec(...)`.
    ℹ NaNs produced

Here the error does not occur:

library(recipes)
library(tidyverse)
library(tidyquant)
library(timetk)

dados <- FANG  %>% select(symbol,date,volume)

#splits <- dados %>% 
#  time_series_split(
#    assess     = "1 months", 
#    cumulative = TRUE
#  )



# Create a recipe object with a timeseries signature step
rec_obj <- recipe(volume ~ ., data = dados) %>%
  
  step_fourier(date,period=365, K = 1)



rec_obj %>% prep() %>% juice() %>% glimpse()

Rows: 4,032
Columns: 5
$ symbol         <fct> FB, FB, FB, FB, FB, FB, FB, FB, FB, FB, FB, FB, FB, F…
$ date           <date> 2013-01-02, 2013-01-03, 2013-01-04, 2013-01-07, 2013…
$ volume         <dbl> 69846400, 63140600, 72715400, 83781800, 45871300, 104…
$ date_sin365_K1 <dbl> 0.2051045, 0.2219215, 0.2386728, 0.2884824, 0.3049212…
$ date_cos365_K1 <dbl> 0.9787401, 0.9750645, 0.9711001, 0.9574852, 0.9523776…
>

Can anyone please help me?

Solution

  • I solve this

    rec_obj <- recipe(volume ~ ., data = training(splits)) %>%

  step_arrange(symbol,date) %>%

  step_fourier(date,period=365, K = 1) %>%

  step_arrange(date,symbol) ```