Skip to main content

A pydantic -> spark schema library

Project description

SparkDantic

codecov PyPI version

1️⃣ version: 0.18.0

✍️ author: Mitchell Lisle

PySpark Model Conversion Tool

This Python module provides a utility for converting Pydantic models to PySpark schemas. It's implemented as a class named SparkModel that extends the Pydantic's BaseModel.

Features

  • Conversion from Pydantic model to PySpark schema.
  • Generate fake data from your model with optional configuration of each field

Usage

Creating a new SparkModel

A SparkModel is a Pydantic model, and you can define one by simply inheriting from SparkModel and defining some fields:

from sparkdantic import SparkModel
from typing import List

class MyModel(SparkModel):
    name: str
    age: int
    hobbies: List[str]

ℹ️ Enums are supported but they must be mixed with either int (IntEnum in Python ≥ 3.10) or str (StrEnum, in Python ≥ 3.11) built-in types:

from enum import Enum

class Switch(int, Enum):
    OFF = 0
    ON = 1

class MyEnumModel(SparkModel):
    switch: Switch

Generating a PySpark Schema

Pydantic has existing models for generating json schemas (with model_json_schema). With a SparkModel you can generate a PySpark schema from the model fields using the model_spark_schema() method:

spark_schema = MyModel.model_spark_schema()

Provides this schema:

StructType([
    StructField('name', StringType(), False),
    StructField('age', IntegerType(), False),
    StructField('hobbies', ArrayType(StringType(), False), False)
])

Generating fake data

Once you've defined a schema you can generate some fake data for your class. Useful for testing purposes as well as for populating development environments with non-production data.

Using the same schema above:

from pyspark.sql import SparkSession
from sparkdantic import SparkModel
from typing import List

class MyModel(SparkModel):
    name: str
    age: int
    hobbies: List[str]
    
spark = SparkSession.builder.getOrCreate()
    
fake_data = MyModel.generate_data(spark)

Using the defaults you'll get fake data that looks something like this (as a Spark DataFrame):

Row(name='0', age=0, hobbies=['0']), 
Row(name='1', age=1, hobbies=['1']), 
Row(name='2', age=2, hobbies=['2']), 
Row(name='3', age=3, hobbies=['3']), 
Row(name='4', age=4, hobbies=['4']), 
Row(name='5', age=5, hobbies=['5']), 
Row(name='6', age=6, hobbies=['6']), 
Row(name='7', age=7, hobbies=['7']), 
Row(name='8', age=8, hobbies=['8']), 
Row(name='9', age=9, hobbies=['9']),

Definitely fake data, but not very useful for replicating the real world. We can provide some more info so that it generates some more realistic figures:

from pyspark.sql import SparkSession
from sparkdantic import SparkModel, ColumnGenerationSpec
from typing import List

class MyModel(SparkModel):
    name: str
    age: int
    hobbies: List[str]
    
spark = SparkSession.builder.getOrCreate()

specs = {
    'name': ColumnGenerationSpec(values=['Bob', 'Alice'], weights=[0.5, 0.5]),
    'age': ColumnGenerationSpec(min_value=20, max_value=65),
    'hobbies': ColumnGenerationSpec(values=['music', 'movies', 'sport'], num_features=2)
}

fake_data = MyModel.generate_data(spark, specs=specs)

Let's breakdown what we've generated:

  • name: We give it a list of two possible values for this field. The weights value determines how often to choose a value. In our case, it chooses one of the two values an equal amount (50% of the rows will be Bob, 50% Alice)
  • age: Choose an int between 20 and 65
  • hobbies: Choose a value from the list at random

There are plenty more options available. Have a look at the library we wrap for this functionality for more examples and information on what you can generate

Row(name='Bob', age=20, hobbies=['music']),
Row(name='Bob', age=21, hobbies=['movies']),
Row(name='Bob', age=22, hobbies=['sport']),
Row(name='Bob', age=23, hobbies=['music']),
Row(name='Bob', age=24, hobbies=['movies']),
Row(name='Alice', age=25, hobbies=['sport']),
Row(name='Alice', age=26, hobbies=['music']),
Row(name='Alice', age=27, hobbies=['movies']),
Row(name='Alice', age=28, hobbies=['sport']),
Row(name='Alice', age=29, hobbies=['music'])

Contributing

Contributions welcome! If you would like to add a new feature / fix a bug feel free to raise a PR and tag me (mitchelllisle) as a reviewer. Please setup your environment locally to ensure all styling and development flow is as close to the standards set in this project as possible. To do this, the main thing you'll need is poetry. You should also run make install-dev-local which will install the pre-commit-hooks as well as install the project locally. PRs won't be accepted without sufficient tests and we will be strict on maintaining a 100% test coverage.

ℹ️ Note that after you have run make install-dev-local and make a commit we run the test suite as part of the pre-commit hook checks. This is to ensure you don't commit code that breaks the tests. This will also try and commit changes to the COVERAGE.txt file so that we can compare coverage in each PR. Please ensure this file is commited with your changes

ℹ️ Versioning: We use bumpversion to maintain the version across various files. If you submit a PR please run bumpversion to the following rules:

  • bumpversion major: If you are making breaking changes (that is, anyone who already uses this library can no longer rely on existing methods / functionality)
  • bumpversion minor: If you are adding functionality or features that maintain existing methods and features
  • bumpversion patch: If you are fixing a bug or making some other small change

Note: ⚠️ You can ignore bumping the version if you like. I periodically do releases of any dependency updates anyway so if you can wait a couple of days for your code to be pushed to PyPi then just submit the change and I'll make sure it's included in the next release. I'll do my best to make sure it's released ASAP after your PR is merged.

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

sparkdantic-0.18.0.tar.gz (10.2 kB view hashes)

Uploaded Source

Built Distribution

sparkdantic-0.18.0-py3-none-any.whl (8.8 kB view hashes)

Uploaded Python 3

Supported by

AWS AWS Cloud computing and Security Sponsor Datadog Datadog Monitoring Fastly Fastly CDN Google Google Download Analytics Microsoft Microsoft PSF Sponsor Pingdom Pingdom Monitoring Sentry Sentry Error logging StatusPage StatusPage Status page