Python SDK

Dataset

SDK Reference: the normalized dataset model

A normalized, immutable view of a loaded tabular dataset. fs.load() returns a Dataset, and it is also what fs.profile(), fs.analyze(), and fs.review() accept as their primary input. The class is a frozen, slotted dataclass so every instance is read-only and safely serializable.

The Dataset type lives in featuresmith.core.dataset. It is not re-exported from the top-level featuresmith package, so import it explicitly:

python

1from featuresmith.core.dataset import Dataset

Dataclass Fields

python

1@dataclass(frozen=True, slots=True)
2class Dataset:
3    dataframe: Any                # pandas or Polars dataframe (memory is not copied)
4    backend: str                  # "pandas" or "polars"
5    schema: DatasetSchema         # columns: tuple[ColumnSchema(name, dtype), ...]
6    metadata: Mapping[str, object]
7    row_count: int                # default 0
8    column_count: int             # default 0
9    dtypes: Mapping[str, str]     # column name -> backend dtype string
10    source: str | None            # original local file path, if any (default None)
11    file_size: int | None         # source file size in bytes, if known (default None)

In practice every field except dataframe, backend, and schema is computed for you during loading, so you normally construct a Dataset via fs.load() or Dataset.from_dataframe() rather than directly.

from_dataframe()

python

1@classmethod
2def from_dataframe(
3    cls,
4    dataframe: Any,
5    *,
6    backend: str,
7    source: str | None = None,
8    file_size: int | None = None,
9    metadata: Mapping[str, object] | None = None,
10) -> Dataset:

Creates a normalized dataset from a pandas or Polars dataframe, inferring the schema and dtype mapping. Used internally by the connectors; the public way to obtain a Dataset is fs.load().

preview(rows=5)

python

1def preview(self, rows: int = 5) -> Any:

Returns the first rows rows of the underlying dataframe (same backend as the source). Raises ValueError when rows is negative.

Example

python

1import featuresmith as fs
2
3dataset = fs.load("train.parquet")
4
5print(dataset.row_count)        # number of rows
6print(dataset.column_count)     # number of columns
7print(dataset.backend)          # "pandas" or "polars"
8print(dataset.source)           # original file path, if loaded from disk
9
10# Inspect the normalized schema and preview rows
11for column in dataset.schema.columns:
12    print(column.name, column.dtype)
13
14print(dataset.preview(5))       # first 5 rows as a dataframe

Python SDK

Dataset

SDK Reference: the normalized dataset model

The Dataset type lives in featuresmith.core.dataset. It is not re-exported from the top-level featuresmith package, so import it explicitly:

python

1from featuresmith.core.dataset import Dataset

Dataclass Fields

python

1@dataclass(frozen=True, slots=True)
2class Dataset:
3    dataframe: Any                # pandas or Polars dataframe (memory is not copied)
4    backend: str                  # "pandas" or "polars"
5    schema: DatasetSchema         # columns: tuple[ColumnSchema(name, dtype), ...]
6    metadata: Mapping[str, object]
7    row_count: int                # default 0
8    column_count: int             # default 0
9    dtypes: Mapping[str, str]     # column name -> backend dtype string
10    source: str | None            # original local file path, if any (default None)
11    file_size: int | None         # source file size in bytes, if known (default None)

from_dataframe()

python

1@classmethod
2def from_dataframe(
3    cls,
4    dataframe: Any,
5    *,
6    backend: str,
7    source: str | None = None,
8    file_size: int | None = None,
9    metadata: Mapping[str, object] | None = None,
10) -> Dataset:

Creates a normalized dataset from a pandas or Polars dataframe, inferring the schema and dtype mapping. Used internally by the connectors; the public way to obtain a Dataset is fs.load().

preview(rows=5)

python

1def preview(self, rows: int = 5) -> Any:

Returns the first rows rows of the underlying dataframe (same backend as the source). Raises ValueError when rows is negative.

Example

python

1import featuresmith as fs
2
3dataset = fs.load("train.parquet")
4
5print(dataset.row_count)        # number of rows
6print(dataset.column_count)     # number of columns
7print(dataset.backend)          # "pandas" or "polars"
8print(dataset.source)           # original file path, if loaded from disk
9
10# Inspect the normalized schema and preview rows
11for column in dataset.schema.columns:
12    print(column.name, column.dtype)
13
14print(dataset.preview(5))       # first 5 rows as a dataframe