Modify add_column() to optionally accept a FeatureType as param

[varadhbhatnagar]

Fix #7142.

Before (Add + Cast):

from datasets import load_dataset, Value
ds = load_dataset("rotten_tomatoes", split="test")
lst = [i for i in range(len(ds))]

ds = ds.add_column("new_col", lst)
# Assigns int64 to new_col by default
print(ds.features)

ds = ds.cast_column("new_col", Value(dtype="uint16", id=None))
print(ds.features)

Before (Numpy Workaround):

from datasets import load_dataset
import numpy as np
ds = load_dataset("rotten_tomatoes", split="test")
lst = [i for i in range(len(ds))]

ds = ds.add_column("new_col", np.array(lst, dtype=np.uint16))
print(ds.features)

After:

from datasets import load_dataset, Value
ds = load_dataset("rotten_tomatoes", split="test")
lst = [i for i in range(len(ds))]
val = Value(dtype="uint16", id=None))
ds = ds.add_column("new_col", lst, feature=val)
print(ds.features)

[varadhbhatnagar]

Requesting review @lhoestq I will also update the docs if this looks good.

[lhoestq]

Cool ! maybe you can rename the argument feature and with type FeatureType ? This way it would work the same way as .cast_column() ?

[varadhbhatnagar]

@lhoestq Since there is no way to get a pyarrow.Schema from a FeatureType, I had to go via Features. How does this look?

[HuggingFaceDocBuilderDev]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.

[varadhbhatnagar]

@lhoestq done!

[varadhbhatnagar]

@lhoestq anything pending on this?