[
https://issues.apache.org/jira/browse/ARROW-15798?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17500701#comment-17500701
]
Dragoș Moldovan-Grünfeld commented on ARROW-15798:
--------------------------------------------------
[~jorisvandenbossche] & [~rokm] I am also bumping a bit into conversion from
double/ float into date.
In R we could do something like this:
{code:r}
# convert a double to date
as.Date(34.56, origin = "1970-01-01")
#> [1] "1970-02-04"
{code}
but then add to it and spill over into the following day
{code:r}
as.Date(34.56, origin = "1970-01-01") + 0.45
#> [1] "1970-02-05"
{code}
due to the fact that (under the hood) the numerical representation is intact
{code:r}
as.numeric(as.Date(34.56, origin = "1970-01-01"))
#> [1] 34.56
{code}
{{date64()}} might be the solution, but the roundtrip to R results in
{{date64()}} being converted as date-time and not date
{code:r}
library(arrow, warn.conflicts = FALSE)
#> See arrow_info() for available features
a <- Array$create(34.56)
(a*86400*1000)$cast(int64())$cast(date64())
#> Array
#> <date64[ms]>
#> [
#> 1970-02-04
#> ]
{code}
In a dplyr pipeline:
{code:r}
library(dplyr, warn.conflicts = FALSE)
library(arrow, warn.conflicts = FALSE)
#> See arrow_info() for available features
df <- tibble::tibble(
a = 34.56
)
df %>%
mutate(b = as.Date(a, origin = "1970-01-01"))
#> # A tibble: 1 × 2
#> a b
#> <dbl> <date>
#> 1 34.6 1970-02-04
df %>%
arrow_table() %>%
mutate(b = as.Date(a, origin = "1970-01-01")) %>%
collect()
#> # A tibble: 1 × 2
#> a b
#> <dbl> <dttm>
#> 1 34.6 1970-02-04 14:26:24
{code}
> [R][C++] Discussion: Plans for date casting from int to support an origin
> option?
> ---------------------------------------------------------------------------------
>
> Key: ARROW-15798
> URL: https://issues.apache.org/jira/browse/ARROW-15798
> Project: Apache Arrow
> Issue Type: Improvement
> Components: C++, R
> Reporter: Dragoș Moldovan-Grünfeld
> Priority: Major
>
> 2 questions:
> * plans to support an origin option for int -> date32 casting?
> * plans to support double -> date32 casting?
> =======================
> Currently the casting from integer to date works, but assumes epoch
> (1970-01-01) as the origin.
> {code:r}
> > a <- Array$create(32L)
> > a$cast(date32())
> Array
> <date32[day]>
> [
> 1970-02-02
> ]
> {code}
> Would it make sense to have an {{origin}} option that would allow the user to
> fine tune the casting? For example, in R the {{base::as.Date()}} function has
> such an argument
> {code:r}
> > as.Date(32, origin = "1970-01-02")
> [1] "1970-02-03"
> {code}
> We have a potential workaround in R (once we support date & duration
> arithmetic), but I was wondering if there might me more general interest for
> this.
> A secondary aspect (as my R example shows) R support casting to date not only
> from integers, but also doubles. Would there be interesting in that? Need be
> I can split this into several tickets.
> Are there any plans in either of these 2 directions?
--
This message was sent by Atlassian Jira
(v8.20.1#820001)
