How to create a time series with tibbletime with parameterized dates?

Question

I want to create a time series with tibbletime for specific dates. I have:

Data_Start<-"2015-09-07 01:55:00 UTC"
Data_End<-"2015-09-10 01:59:00 UTC"

and I want to create a timeseries, with minute samples, like with:

create_series(2015-09-07 + 01:55:00 ~ 2015-09-10 + 01:59:00,1~M)

The parameters should be a time_formula, described on page 17 here: https://cran.r-project.org/web/packages/tibbletime/tibbletime.pdf

This works, but I cannot pass the parameters like:

create_series(Data_Start~Data_End,1~M)

Tried out already different things for converting the string, but nothing worked so far :(

Answer 1

Author of tibbletime here. An issue on GitHub was raised about this recently. The solution is to use rlang::new_formula() to pre-construct the formula. We also need a special helper function that can handle adding the + in the formula if using POSIXct dates.

Here is the helper:

# Time formula creator
# Can pass character, Date, POSIXct
create_time_formula <- function(lhs, rhs) {

  if(!inherits(lhs, c("character", "Date", "POSIXct"))) {
    stop("LHS must be a character or date")
  }
  if(!inherits(rhs, c("character", "Date", "POSIXct"))) {
    stop("RHS must be a character or date")
  }

  if(inherits(lhs, "Date")) {
    lhs <- as.character(lhs)
  } else if (inherits(lhs, "POSIXct")) {
    lhs <- gsub(" ", " + ", lhs)
  }

  if(inherits(rhs, "Date")) {
    rhs <- as.character(rhs)
  } else if (inherits(rhs, "POSIXct")) {
    rhs <- gsub(" ", " + ", rhs)
  }

  rlang::new_formula(lhs, rhs)
}

Use the helper function with date versions of your start and end dates

Data_Start<- as.POSIXct("2015-09-07 01:55:00")
Data_End  <- as.POSIXct("2015-09-10 01:59:00")

time_formula <- create_time_formula(Data_Start, Data_End)

create_series(time_formula, 1~M, tz = "UTC")

Produces:

# A time tibble: 4,325 x 1
# Index: date
                  date
                <dttm>
 1 2015-09-07 01:55:00
 2 2015-09-07 01:56:00
 3 2015-09-07 01:57:00
 4 2015-09-07 01:58:00
 5 2015-09-07 01:59:00
 6 2015-09-07 02:00:00
 7 2015-09-07 02:01:00
 8 2015-09-07 02:02:00
 9 2015-09-07 02:03:00
10 2015-09-07 02:04:00
# ... with 4,315 more rows

In a future release of tibbletime I will likely include a more robust verson of the create_time_formula() helper function for this case.

Update: tibbletime 0.1.0 has been released, and a more robust implementation allows for directly using variables in the formula. Additionally, each side of the formula must be a character or an object of the same class as the index now (i.e. 2013 ~ 2014 should be "2013" ~ "2014").

library(tibbletime)

Data_Start<- as.POSIXct("2015-09-07 01:55:00")
Data_End  <- as.POSIXct("2015-09-10 01:59:00")

create_series(Data_Start ~ Data_End, "1 min")
#> # A time tibble: 4,325 x 1
#> # Index: date
#>    date               
#>    <dttm>             
#>  1 2015-09-07 01:55:00
#>  2 2015-09-07 01:56:00
#>  3 2015-09-07 01:57:00
#>  4 2015-09-07 01:58:00
#>  5 2015-09-07 01:59:00
#>  6 2015-09-07 02:00:00
#>  7 2015-09-07 02:01:00
#>  8 2015-09-07 02:02:00
#>  9 2015-09-07 02:03:00
#> 10 2015-09-07 02:04:00
#> # ... with 4,315 more rows

Answer 2

I have created timeseries with multiple seasonality using forecast()package between the mentioned time with minutes as frequency. Seasonal periods differs with your requirement and data length

library(forecast)
Data_Start<-as.POSIXct("2015-09-07 01:55:00 UTC")
Data_End<-as.POSIXct("2015-09-10 01:59:00 UTC")

df = data.frame(tt = seq.POSIXt(Data_Start,Data_End,"min"),
                val = sample(1:40,4325,replace = T),stringsAsFactors = F)

# Seasonality Hourly, Daily
mts = msts(df$val,seasonal.periods = c(60,1440),start = Data_Start)
# Seasonality Hourly, Daily, Weekly
mts = msts(df$val,seasonal.periods = c(60,1440,10080),start = Data_Start)