Error in step_fourier from timetk package

Question

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?

Can you please include the error message, or even better, provide an example using built-in data sets — EmilHvitfeldt, Nov 28 '22 at 20:33
There was no error message, when I executed the exact same code. Maybe you need to update your libraries. Also you need to load the library rsample for the use of the training function. Maybe there was some library overlap too — Leonhard Geisler, Nov 29 '22 at 08:55
Hello, thanks for the contribution. I tried it on my local machine and through Databricks, it gives the same error. — Mairon Chaves, Nov 29 '22 at 12:39

score 0 · Accepted Answer · answered Jan 10 '23 at 13:16

0

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) ```

answered Jan 10 '23 at 13:16

Mairon Chaves

13
3

Error in step_fourier from timetk package

1 Answers1