Source code for maxframe.liteframe.initializer

# Copyright 1999-2026 Alibaba Group Holding Ltd.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#      http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.

import pandas as pd

from maxframe.liteframe.core import LiteFrame as _LiteFrame
from maxframe.serialization.serializables import SerializableMeta
from maxframe.tensor.core import TENSOR_TYPE


class InitializerMeta(SerializableMeta):
    def __instancecheck__(cls, instance):
        return isinstance(
            instance, (cls.__base__,) + getattr(cls, "_allow_data_type_", ())
        )


[docs] class LiteFrame(_LiteFrame, metaclass=InitializerMeta): """A performance-dedicated, lazy, distributed tabular data object with no index semantics. LiteFrame provides a pandas-looking DataFrame API for large-scale data processing on MaxCompute. Unlike a traditional DataFrame, LiteFrame has no row index: all operations are column-based, single-column selection returns a LiteFrame (never a Series), and there is no implicit label alignment between independent frames. Arrow dtypes are used as first-class citizens so that integer columns with nulls stay nullable integer rather than being silently promoted to float. By removing index overhead and embedding element-wise expressions in a small set of fusible operators, LiteFrame achieves lower planning cost and higher execution throughput. Operations are lazy — they build a computation graph that is executed on demand via ``.execute()``. Row order is preserved through execution unless an operator explicitly reorders rows. Parameters ---------- data : dict, pandas.DataFrame, tensor, or LiteFrame Input data. A ``dict`` or ``pandas.DataFrame`` is converted internally to an Arrow-backed LiteFrame. A tensor is wrapped as a single-column (or multi-column) frame. An existing LiteFrame is returned as-is. columns : list of str, optional Column labels to use. When ``data`` is a dict or DataFrame, this selects and orders the output columns. dtype : dtype, optional Data type to force on all columns. chunk_size : int, optional Row-axis chunk size for distributed execution. gpu : bool, optional Whether to use GPU execution. See Also -------- maxframe.liteframe.read_csv : Read a CSV file into a LiteFrame. maxframe.liteframe.read_parquet : Read a Parquet file into a LiteFrame. maxframe.liteframe.concat : Concatenate LiteFrames along the row axis. maxframe.liteframe.LiteFrame.from_tensor : Create a LiteFrame from a MaxFrame tensor. Examples -------- >>> import maxframe.liteframe as mlf >>> lf = mlf.LiteFrame({'a': [1, 2, 3], 'b': [4, 5, 6]}) >>> lf.execute() a b 0 1 4 1 2 5 2 3 6 """
[docs] def __init__( self, data=None, columns=None, dtype=None, chunk_size=None, gpu=None, ): # Lazy import to avoid circular dependency with datasource module from maxframe.liteframe.datasource.from_local import from_local_df from maxframe.liteframe.datasource.from_tensor import liteframe_from_tensor if isinstance(data, _LiteFrame): super().__init__(data.data) return if isinstance(data, TENSOR_TYPE): tensor = data # LiteFrame is only chunked along the row axis; when a chunk size is # given, rechunk the rows of the tensor accordingly. if chunk_size is not None: tensor = tensor.rechunk({0: chunk_size}) lf = liteframe_from_tensor(tensor, columns=columns, gpu=gpu) if dtype is not None: lf = lf.astype(dtype) super().__init__(lf.data) return if isinstance(data, dict): pdf = pd.DataFrame(data, columns=columns, dtype=dtype) elif isinstance(data, pd.DataFrame): pdf = data.copy() if columns is not None: pdf = pdf[columns] elif data is None: pdf = pd.DataFrame(columns=columns or [], dtype=dtype) else: pdf = pd.DataFrame(data, columns=columns, dtype=dtype) lf = from_local_df(pdf, chunk_size=chunk_size, gpu=gpu) super().__init__(lf.data)