consider using {tsibble} about incidence2 HOT 6 CLOSED

Cheers for the feedback - really useful!

• Re: tsibble - the "alpha alpha" version of incidence2 was actually built on top of tsibble. It is a really nice package! However because each row of inputs we deal with can be considered a unique observation there was little need for the key used in tsibble. It felt like we were adding unnecessary dependencies for a small subset of the functionality. At this stage, with dplyr and tibble still refining how best to extend their functions/classes, I'm wary of relying on an intermediate package in the implementation. Having said that we're definitely not unwilling to change. The underlying implementation is such that we could swap out for tsibble in future if everyone felt that was the way to go!

• In regards to aweek, tsibble is implemented in a way which lets you use the aweek package for grouping should you wish but I agree it would be a useful function to be implemented directly. If it helps I can dig out my code that shows you how to use aweek with tsibble.

• On naming - I will definitely look in to that. @thibautjombart and I were only discussing this morning how now was the time to change any names from the original package.

from incidence2.

@aspina7 here you go. Do you mind if I post this on the tsibble issue as well in case others are interested?

library(outbreaks)
library(dplyr, warn.conflicts = FALSE)
library(tsibble)
library(aweek)
library(vctrs)
library(lubridate, warn.conflicts = FALSE)

# bits for tsibble --------------------------------------------------------
index_valid.aweek <- function(x) {
  TRUE
}

interval_pull.aweek <- function(x) {
  wk <- as.double((as_date(x) - as_date("1969-12-29")) / 7)
  nweeks <- tsibble:::gcd_interval(wk)
  new_interval(week = nweeks)
}

vec_cast.double.aweek <- function(x, to, ...) {
  as.double((as_date(x) - as_date("1969-12-29")) / 7)
}


# load data ---------------------------------------------------------------
dat_raw <- ebola_sim_clean$linelist

dat_ts <-
  dat_raw %>%
  as_tsibble(key = case_id, index = date_of_onset)


# aweek epiweek version ---------------------------------------------------
dat_ts %>%
  group_by(hospital, outcome) %>%
  index_by(epiweek = ~ date2week(., week_start = "Sunday", floor_day = TRUE)) %>%
  group_by(hospital, outcome) %>%
  summarise(count = n())
#> # A tsibble: 849 x 4 [1W]
#> # Key:       hospital, outcome [18]
#> # Groups:    hospital [6]
#>    hospital           outcome epiweek    count
#>    <fct>              <fct>   <aweek>    <int>
#>  1 Connaught Hospital Death   2014-W22-1     3
#>  2 Connaught Hospital Death   2014-W23-1     3
#>  3 Connaught Hospital Death   2014-W24-1     1
#>  4 Connaught Hospital Death   2014-W25-1     6
#>  5 Connaught Hospital Death   2014-W26-1     4
#>  6 Connaught Hospital Death   2014-W27-1     1
#>  7 Connaught Hospital Death   2014-W28-1     4
#>  8 Connaught Hospital Death   2014-W29-1     6
#>  9 Connaught Hospital Death   2014-W30-1     4
#> 10 Connaught Hospital Death   2014-W31-1    12
#> # … with 839 more rows

^{Created on 2020-07-03 by the reprex package (v0.3.0)}

from incidence2.

awesome thanks!
Nay dont mind at all - go for it.

(will close this issue)

from incidence2.

The underlying implementation is such that we could swap out for tsibble in future if everyone felt that was the way to go!

Fair - makes sense the tsibble key (while functional) has been a bit annoying at times!

If it helps I can dig out my code that shows you how to use aweek with tsibble.

oo yes please that would be awesome!

from incidence2.

hey @tjtnew just circling back to this, based on the other discussion above and the fix done over in tsibble. Worth switching incidence2 to use lubridate rather than aweek?

from incidence2.

hey @aspina7 I was actually looking at this this, will summarise where I'm at below:

• I'm trying to work towards the v1.0 release of incidence2 and a major part of this is standardising how the intervals / date groups work. Currently, only the week groupings (due to aweek) generate a proper s3 class that you can work with independently from the incidence dataframes.
• Short term my plan is to standardise all of the date groupings (week, month, quarter and year) and implement them with the vctrs package. This will mean I move away from aweek but may not necessitate using lubridate (my current rough implementation only uses the wday function from there which I think I can avoid by converting Dates to POSIXlt and getting the weekday from there). The months and quarters groupings will then be implemented in a way similar to the implementation here https://github.com/DavisVaughan/datea.
• Longer term, it looks like Davis Vaughan is working on a very cool package called clock. Once this is released I think all our date grouping functionality will be replaced by this. It looks like it's going to be a really well thought out packages for dealing with dates and datetimes!

from incidence2.