Temporal

[Polars Documentation]

OpenDP supports some manipulation of dates and times, which can be useful in predicates and grouping functions.

import polars as pl

import opendp.prelude as dp
dp.enable_features("contrib")
# Fetch and unpack the data.
![ -e ../sample_FR_LFS.csv ] || ( curl 'https://github.com/opendp/dp-test-datasets/blob/main/data/sample_FR_LFS.csv.zip?raw=true' --location --output sample_FR_LFS.csv.zip; unzip sample_FR_LFS.csv.zip -d ../ )

lf_dates = (
    pl.scan_csv("../sample_FR_LFS.csv", ignore_errors=True)
    # prepare the data with some expressions that are not yet supported in OpenDP
    .select(DATE=pl.concat_str("REFYEAR", pl.col.QUARTER * 3, pl.lit("01"), separator="-"))
)

context = dp.Context.compositor(
    data=lf_dates,
    privacy_unit=dp.unit_of(contributions=36),
    privacy_loss=dp.loss_of(epsilon=1.0, delta=1e-7),
    split_evenly_over=1,
)

Date/Time Components

• Date expressions (can be applied to pl.Date and pl.Datetime dtypes)

  • .dt.year

  • .dt.iso_year

  • .dt.quarter

  • .dt.month

  • .dt.week

  • .dt.weekday

  • .dt.day

  • .dt.ordinal_day

• Time expressions (can be applied to pl.Time and pl.Datetime dtypes)

  • .dt.hour

  • .dt.minute

  • .dt.second

  • .dt.millisecond

  • .dt.microsecond

  • .dt.nanosecond

An example of their use can be seen below, where a string column is parsed into dates, and then year and month components are retrieved from the dates.

query = (
    context.query()
    .with_columns(pl.col.DATE.str.to_date(format=r"%Y-%m-%d"))
    .with_columns(YEAR=pl.col.DATE.dt.year(), MONTH=pl.col.DATE.dt.month())
    .group_by("YEAR", "MONTH")
    .agg(dp.len())
)
query.release().collect().sort("YEAR", "MONTH")

shape: (40, 3)

YEAR	MONTH	len
i32	i8	u32
2004	3	4209
2004	6	4179
2004	9	4123
2004	12	4061
2005	3	4136
…	…	…
2012	12	6424
2013	3	6300
2013	6	5905
2013	9	5618
2013	12	5702