DST temporal feature from timestamp using polars

I'm migrating code to polars from pandas. I have time-series data consisting of a timestamp and value column and I need to compute a bunch of features. i.e.

df = pl.DataFrame({
    "timestamp": pl.date_range(
        datetime(2017, 1, 1),
        datetime(2018, 1, 1),
        timedelta(minutes=15),
        time_zone="Australia/Sydney",
        time_unit="ms", eager=True),
        })
    value = np.random.normal(0, 1, len(df))
    df = df.with_columns([pl.Series(value).alias("value")])

I need to generate a column containing an indicator if the timestamp is standard or daylight time. I'm currently using apply because as far as I can see the isn't a Temporal Expr, i.e. my current code is

def dst(timestamp:datetime):
    return int(timestamp.dst().total_seconds()!=0)

df = df.with_columns(pl.struct(["timestamp"]).apply(lambda x: dst(**x)).alias("dst"))

(this uses a trick that effectively checks if the tzinfo.dst(dt) offset is zero or not)

Is there a (fast) way of doing this using polars expressions rather than (slow) apply?

You can exploit strftime for this.

(
    df
        .with_columns(
            dst=pl.when(pl.col('timestamp').dt.strftime("%Z").str.contains("(DT$)"))
            .then(True)
            .otherwise(False)
            )
)

It relies on the local time zone ending in "DT" to determine the dst status. That works here and would work for US time zones (ie EST/EDT, CST/CDT, etc) but examples that wouldn't work are numerous.

Alternatively you could use the utc offset but it's a lot more convoluted.

(
    df
        .with_columns(
            tzoff=pl.col('timestamp').dt.strftime("%z").cast(pl.Int64())
            )
    .join(
        df
            .select(
                tzoff=pl.col('timestamp').dt.strftime("%z").cast(pl.Int64())
                )
            .unique('tzoff')
            .sort('tzoff')
            .with_columns(
                dst=pl.lit([False, True])
                ), 
        on='tzoff')
    .drop('tzoff')
)

This one assumes that the timezone only has 2 offsets and that the smaller of the two is standard time and the bigger one is daylight savings.

With polars>=0.18.5 the following works

df = df.with_columns((pl.col("timestamp").dt.dst_offset()==0).cast(pl.Int32).alias("dst"))