groupby aggregate functions about polars HOT 6 CLOSED

Such as:

• .mean()
• .quantile()
• .median()
• .first()
• .last()
• .min()
• .max()
• .sum()
• .nunique()

from polars.

Status:

• mean
• median
• quantile
• first
• last
• min
• max
• sum
• nunique
• agg to series
• return group indexes

from polars.

I just took this for a spin. I wondered. The there a reason for the .select().sum() syntax as opposed to the .agg() in pandas?

from polars.

I just took this for a spin. I wondered. The there a reason for the .select().sum() syntax as opposed to the .agg() in pandas?

No specific reason. The python api closely resembles the Rust api and isn't per se meant to be a copy of Pandas (however familiarity is beneficial).

We could default to all selected columns by default and only aggregate the columns that are summable. That would be a more ergonomic default. groupby("foo").sum()

from polars.

In pandas right now I can do;

df.groupby('user_id').agg(n_sess=('session', 'nunique'), avg_session_len=('session_time', 'mean'))

After grouping I can calculate multiple aggregations and I can be specific. I can specfiy here that I want the number of unique sessions and the average session time. How might that work here?

from polars.

Right! We cannot yet. 😄 This was already requested in #75 . It's coming. I think that we can also support the same syntax on the Python side

Edit:
agg syntax and multiple aggregations added in cb85513

python

 gb_df = df.groupby("a").agg({"b": ["sum", "min"], "c": "count"})

rust

df.groupby("a")?.agg(&[("b", &["n_unique", "sum", "min"]), ("c", &["count"])])

from polars.