PDL::DateTime

[zmughal]

Pandas:

df.time_col.weekday

R

?Date
?POSIXt
?weekdays
?seq.Date

[kmx]

Hi Zaki,

do you have some particular ideas how PDL::DateTime should be implemented?

I am using my module (not released as it is still an ugly hack) that implements PDL::SerialDate (more or less the serial date from matlab). It basically is a PDL (Long or LongLong) with overloaded stringification (to print date) and some special constructors like new_from_ymd() and few extra methods. Now it handles only dates but I can imagine (although not tested) that with underlying PDL of type Double it might also handle date+time.

[zmughal]

I was thinking about storing the POSIX epoch time in a subclass of PDL::Long. It would be like R's POSIXct. And it will support methods that extract out the weekday (returning a PDL::Factor subclass) or month (another PDL::Factor subclass). There may also be a method that returns a PDL::SV full of DateTime(::Tiny) instances.

[kmx]

I have started some experiments. I am sub-classing PDL::LongLong internally keeping number of microseconds (positive or negative) since 1970-01-01. Microseconds might come handy when dealing with high-frequency data. LongLong enables us to handle date time range 0001-01-01 00:00:00.000000 .. 9999-12-31 23:59:59.999999 with microsecond precision.

BTW LongLong + nanosecond precision would limit us to date range 1677-09-21T00:12:44 .. 2262-04-11T23:47:15

Ad converting date time values to "some" instances - DateTime is not a good option performance-wise, Time::Moment is very promising.

[kmx]

Hi Zaki,

FYI my hacking on PDL::DateTime is here

• https://github.com/kmx/pdl-datetime/blob/master/lib/PDL/DateTime.pm
• https://github.com/kmx/pdl-datetime/blob/master/lib/PDL/DateTime.pod

Any comments are welcome.

[zmughal]

Cool! I'll definitely follow along!

[kmx]

I have finish kind of a initial PDL::DateTime version (see my repo). It only needs PDL core with longlong-double fix

[kmx]

Zaki,

I want to ask if your original plan was to have PDL::DateTime a part of Data-Frame distribution or a separate module. Is it OK for you if I release PDL::DateTime as my module to CPAN?

My plan is to wait for PDL-2.014 (longlong fixes, you know) and then release PDL::DateTime + also a new versions of PDL::IO::DBI and PDL::IO::CSV which will have a support for automatic conversion of DATETIME data to PDL::DateTime piddles.

--kmx

[zmughal]

On 2015-10-05 at 03:56:20 -0700, kmx wrote:

Zaki,

I want to ask if your original plan was to have PDL::DateTime a part of Data-Frame distribution or a separate module. Is it OK for you if I release PDL::DateTime as my module to CPAN?

My plan is to wait for PDL-2.014 (longlong fixes, you know) and then release PDL::DateTime + also a new versions of PDL::IO::DBI and PDL::IO::CSV which will have a support for automatic conversion of DATETIME data to PDL::DateTime piddles.

A separate module is perfectly fine. I only made reference to it here
because I was going through the functionality of Pandas from Python
which includes its own time series specific data structure.

Cheers,
-- Zaki Mughal

--kmx

---

Reply to this email directly or view it on GitHub:
#18 (comment)

[zmughal]

No longer needed as @kmx has released https://metacpan.org/pod/PDL::DateTime.