pandas-dataclasses 0.4.0

pandas data creation made easy by dataclass

pandas-dataclasses

pandas data creation made easy by dataclass

Overview

pandas-dataclass makes it easy to create pandas data (Series and DataFrame) by Python's dataclass that enables to specify their data types, attributes, and names:

Click to see all imports

from dataclasses import dataclass
from pandas_dataclasses import AsDataFrame, Data, Index

@dataclass
class Weather(AsDataFrame):
    """Weather information."""

    year: Index[int]
    month: Index[int]
    temp: Data[float]
    humid: Data[float]


df = Weather.new(
    [2020, 2020, 2021, 2021, 2022],
    [1, 7, 1, 7, 1],
    [7.1, 24.3, 5.4, 25.9, 4.9],
    [65, 89, 57, 83, 52],
)

where df will become a DataFrame object like:

            temp  humid
year month
2020 1       7.1   65.0
     7      24.3   89.0
2021 1       5.4   57.0
     7      25.9   83.0
2022 1       4.9   52.0

Features

• Type specification of pandas indexes and data
• Metadata storing in pandas data attributes
• Support for hierarchical index and columns
• Support for full dataclass features
• Support for static type check by Pyright (Pylance)

Installation

pip install pandas-dataclasses

How it works

pandas-dataclasses provides you the following features:

• Type hints for dataclass fields (Attr, Data, Index, Name) to specify index(es), data, attributes, and names of pandas data
• Mix-in classes for dataclasses (As, AsDataFrame, AsSeries) to create pandas data by a classmethod (new) that takes the same arguments as dataclass initialization

When you call new, it will first create a dataclass object and then create a Series or DataFrame object from the dataclass object according the type hints and values in it. In the example above, df = Weather.new(...) is thus equivalent to:

obj = Weather([2020, ...], [1, ...], [7.1, ...], [65, ...])
df = asdataframe(obj)

where asdataframe is a conversion function. pandas-dataclasses does not touch the dataclass object creation itself; this allows you to fully customize your dataclass before conversion by the dataclass features (field, __post_init__, ...).

Basic usage

DataFrame creation

As shown in the example above, a dataclass that has the AsDataFrame mix-in will create DataFrame objects:

Click to see all imports

from dataclasses import dataclass
from pandas_dataclasses import AsDataFrame, Data, Index

@dataclass
class Weather(AsDataFrame):
    """Weather information."""

    year: Index[int]
    month: Index[int]
    temp: Data[float]
    humid: Data[float]


df = Weather.new(...)

where fields typed by Index are "index fields", each value of which will become an index or a part of a hierarchical index of a DataFrame object. Fields typed by Data are "data fields", each value of which will become a data column of a DataFrame object. Fields typed by other types are just ignored in the DataFrame creation.

Each data or index will be cast to the data type specified in a type hint like Index[int]. Use Any or None (like Index[Any]) if you do not want type casting. See data typing rules for more examples.

By default, a field name (i.e. an argument name) is used for the name of corresponding data or index. See custom data/index naming if you want customization.

Series creation

A dataclass that has the AsSeries mix-in will create Series objects:

Click to see all imports

from dataclasses import dataclass
from pandas_dataclasses import AsSeries, Data, Index

@dataclass
class Temperature(AsSeries):
    """Temperature information."""

    year: Index[int]
    month: Index[int]
    temp: Data[float]


ser = Temperature.new(...)

Unlike AsDataFrame, the second and subsequent data fields are ignored in the Series creation. Other rules are the same as for the DataFrame creation.

Advanced usage

Metadata storing

Fields typed by Attr are "attribute fields", each value of which will become an item of attributes (attrs) of a DataFrame or a Series object:

Click to see all imports

from dataclasses import dataclass
from pandas_dataclasses import AsDataFrame, Attr, Data, Index

@dataclass
class Weather(AsDataFrame):
    """Weather information."""

    year: Index[int]
    month: Index[int]
    temp: Data[float]
    humid: Data[float]
    loc: Attr[str] = "Tokyo"
    lon: Attr[float] = 139.69167
    lat: Attr[float] = 35.68944

In this example, Weather.new(...).attrs will become like:

{"loc": "Tokyo", "lon": 139.69167, "lat": 35.68944}

Custom naming

The name of data, index, or attribute can be explicitly specified by adding a string annotation to the corresponding type:

Click to see all imports

from dataclasses import dataclass
from typing import Annotated as Ann
from pandas_dataclasses import AsDataFrame, Attr, Data, Index

@dataclass
class Weather(AsDataFrame):
    """Weather information."""

    year: Ann[Index[int], "Year"]
    month: Ann[Index[int], "Month"]
    temp: Ann[Data[float], "Temperature (deg C)"]
    humid: Ann[Data[float], "Humidity (%)"]
    loc: Ann[Attr[str], "Location"] = "Tokyo"
    lon: Ann[Attr[float], "Longitude (deg)"] = 139.69167
    lat: Ann[Attr[float], "Latitude (deg)"] = 35.68944

In this example, Weather.new(...) and its attributes will become like:

            Temperature (deg C)  Humidity (%)
Year Month
2020 1                      7.1          65.0
     7                     24.3          89.0
2021 1                      5.4          57.0
     7                     25.9          83.0
2022 1                      4.9          52.0

{"Location": "Tokyo", "Longitude (deg)": 139.69167, "Latitude (deg)": 35.68944}

Adding dictionary annotations to data fields will create DataFrame objects with hierarchical columns, where dictionary keys will become the names of column levels and dictionary values will become the names of columns:

Click to see all imports

from dataclasses import dataclass
from typing import Annotated as Ann
from pandas_dataclasses import AsDataFrame, Data, Index

def name(stat: str, cat: str) -> dict[str, str]:
    return {"Statistic": stat, "Category": cat}


@dataclass
class Weather(AsDataFrame):
    """Weather information."""

    year: Ann[Index[int], "Year"]
    month: Ann[Index[int], "Month"]
    temp_avg: Ann[Data[float], name("Temperature (degC)", "Average")]
    temp_max: Ann[Data[float], name("Temperature (degC)", "Maximum")]
    wind_avg: Ann[Data[float], name("Wind speed (m/s)", "Average")]
    wind_max: Ann[Data[float], name("Wind speed (m/s)", "Maximum")]

In this example, Weather.new(...) will become like:

Statistic  Temperature (degC)         Wind speed (m/s)
Category              Average Maximum          Average Maximum
Year Month
2020 1                    7.1    11.1              2.4     8.8
     7                   24.3    27.7              3.1    10.2
2021 1                    5.4    10.3              2.3    10.7
     7                   25.9    30.3              2.4     9.0
2022 1                    4.9     9.4              2.6     8.8

For the Series creation, a field typed by Name is a "name field", whose value will become the name of a Series object. This is useful for dynamic naming. See also naming rules for more details and examples.

Custom pandas factory

A custom class can be specified as a factory for the Series or DataFrame creation by As, the generic version of AsDataFrame and AsSeries. Note that the custom class must be a subclass of either pandas.Series or pandas.DataFrame:

Click to see all imports

import pandas as pd
from dataclasses import dataclass
from pandas_dataclasses import As, Data, Index

class CustomSeries(pd.Series):
    """Custom pandas Series."""

    pass


@dataclass
class Temperature(As[CustomSeries]):
    """Temperature information."""

    year: Index[int]
    month: Index[int]
    temp: Data[float]


ser = Temperature.new(...)
isinstance(ser, CustomSeries)  # True

Appendix

Data typing rules

The data type (dtype) of data/index is inferred from the first Data/Index type of the corresponding field. The following table shows how the data type is inferred:

Click to see all imports

from typing import Any, Annotated as Ann, Literal as L
from pandas_dataclasses import Data

Type hint	Inferred data type
Data[Any]	None (no type casting)
Data[None]	None (no type casting)
Data[int]	numpy.dtype("i8")
Data[numpy.int32]	numpy.dtype("i4")
Data[L["datetime64[ns]"]]	numpy.dtype("<M8[ns]")
Data[L["category"]]	pandas.CategoricalDtype()
Data[int] | str	numpy.dtype("i8")
Data[int] | Data[float]	numpy.dtype("i8")
Ann[Data[int], "spam"]	numpy.dtype("i8")
Data[Ann[int, "spam"]]	numpy.dtype("i8")

Naming rules

The name of data/index is determined by the following rules:

1. If a name field exists, its value will be preferentially used (Series creation only)
2. If a data/index field is annotated, the first annotation in the first Data/Index type will be used
3. Otherwise, the field name (i.e. argument name) will be used

The following table shows how the name is inferred in the case of 2 and 3:

Click to see all imports

from typing import Any, Annotated as Ann
from pandas_dataclasses import Data

Type hint	Inferred name
Data[Any]	Field name
Ann[Data[Any], "spam"]	"spam"
Ann[Data[Any], "spam"]	"spam"
Ann[Data[Any], "spam", "ham"]	"spam"
Ann[Data[Any], "spam"] | Ann[str, "ham"]	"spam"
Ann[Data[Any], "spam"] | Ann[Data[float], "ham"]	"spam"
Ann[Data[Any], {"0": "spam", "1": "ham"}]	("spam", "ham")

Development roadmap

Release version	Features
v0.4.0	Support for hierarchical column
v0.5.0	Support for dynamic naming of indexes and data
v1.0.0	Initial major release (freezing public features until v2.0.0)