Python-Polars: group_by a column have the remaining columns be combined into a struct field?

I have not found a way to group_by() a column and have a struct field for the new column.Here is a snippet of the dataframe:

Essentially what I am trying to do is to group by Census_Tract and GEOID_Census. The columns table_B24012 and value should be combined into one column.

In other words I would like the data to be a struct object such that for each GEOID_Census I have one row value with potential a dictionary structure built into the new column. It would look something like this

If there is a more efficient way to store it in the column, I am open to other ways. I am just trying to condense the information.

I have tried casting the two columns to a list and then struct, as well as just to a dictionary. Nothing has helped. Any ideas? Thanks in advance!

It sounds like you want something similar to a .pivot()

Data

# increase defaults
pl.Config(fmt_table_cell_list_len=10, fmt_str_lengths=100)

df = pl.from_repr("""
┌────────────────┬──────────────┬──────────────┬─────────┐
│ Census_Tract   ┆ GEOID_Census ┆ table_B24012 ┆ value   │
│ ---            ┆ ---          ┆ ---          ┆ ---     │
│ str            ┆ str          ┆ str          ┆ f64     │
╞════════════════╪══════════════╪══════════════╪═════════╡
│ Census Tract 9 ┆ 04001942600  ┆ B24012_001E  ┆ 11111.0 │
│ Census Tract 9 ┆ 04001942700  ┆ B24012_002E  ┆ 22222.0 │
│ Census Tract 9 ┆ 04001942600  ┆ B24012_002E  ┆ 33333.0 │
│ Census Tract 9 ┆ 04001942700  ┆ B24012_001E  ┆ 44444.0 │
│ Census Tract 9 ┆ 04001944100  ┆ B24012_001E  ┆ 55555.0 │
└────────────────┴──────────────┴──────────────┴─────────┘
""")

Pivot

df.pivot("table_B24012", index=["Census_Tract", "GEOID_Census"])

shape: (3, 4)
┌────────────────┬──────────────┬─────────────┬─────────────┐
│ Census_Tract   ┆ GEOID_Census ┆ B24012_001E ┆ B24012_002E │
│ ---            ┆ ---          ┆ ---         ┆ ---         │
│ str            ┆ str          ┆ f64         ┆ f64         │
╞════════════════╪══════════════╪═════════════╪═════════════╡
│ Census Tract 9 ┆ 04001942600  ┆ 11111.0     ┆ 33333.0     │
│ Census Tract 9 ┆ 04001942700  ┆ 44444.0     ┆ 22222.0     │
│ Census Tract 9 ┆ 04001944100  ┆ 55555.0     ┆ null        │
└────────────────┴──────────────┴─────────────┴─────────────┘

Struct

It's easy to create a struct from there if desired:

(df.pivot("table_B24012", index=["Census_Tract", "GEOID_Census"])
   .select(
       "Census_Tract", 
       "GEOID_Census", 
       table_B24012 = pl.struct(pl.exclude("Census_Tract", "GEOID_Census"))
   )
)

shape: (3, 3)
┌────────────────┬──────────────┬───────────────────┐
│ Census_Tract   ┆ GEOID_Census ┆ table_B24012      │
│ ---            ┆ ---          ┆ ---               │
│ str            ┆ str          ┆ struct[2]         │
╞════════════════╪══════════════╪═══════════════════╡
│ Census Tract 9 ┆ 04001942600  ┆ {11111.0,33333.0} │
│ Census Tract 9 ┆ 04001942700  ┆ {44444.0,22222.0} │
│ Census Tract 9 ┆ 04001944100  ┆ {55555.0,null}    │
└────────────────┴──────────────┴───────────────────┘