PyGlove

0.4.5 · active · verified Wed Apr 15

PyGlove is a general-purpose Python library for manipulating Python objects, introducing symbolic object-oriented programming (SOOP). It facilitates meta-program writing and is used in complex machine learning scenarios like AutoML, evolutionary computing, and daily programming tasks by providing a mutable symbolic object model and a rich set of object manipulation operations. It is currently at version 0.4.5 and actively maintained with regular updates.

Warnings

breaking When overriding `pg.Object.__init__`, it is now mandatory to apply the `@pg.explicit_method_override` decorator. Failing to do so will result in an error, as it prevents accidental overrides of core PyGlove-managed methods.
Fix: Add the `@pg.explicit_method_override` decorator to any overridden `__init__` method in `pg.Object` subclasses.
deprecated Several internal APIs were renamed. `pg.Object.schema` became `pg.Object.__schema__`, `pg.Object.type_name` became `pg.Object.__type_name__`, and `pg.Object.serialization_key` moved to `pg.JSONConvertible.__serialization_key__`. Similarly, `pg.Functor.signature` was renamed to `pg.Functor.__signature__`. Additionally, `pg.ContextualValue`, `pg.symbolic.GetAttributeContext`, and related methods were removed.
Fix: Update usage to the new private API names (e.g., `__schema__`) or avoid direct use of removed internal components. Refer to release notes for precise renamings.
gotcha PyGlove operates on a mutable programming paradigm, which can introduce complexities and limitations, especially when refactoring existing code. Adapting models and trainers to conform to PyGlove's symbolic object constraints, particularly for generic subclassed models, can be more challenging than expected, as symbolic nodes constructed by initialization parameters might not behave as anticipated.
Fix: Carefully review PyGlove's documentation on symbolic object-oriented programming (SOOP) and mutation. Be prepared for potential refactoring and validation efforts when integrating with complex or highly customized object hierarchies.
deprecated The `pg.members` function and its alias `pg.schema` were replaced by symbolic member declaration using field annotations as the recommended approach for defining symbolic fields within classes.
Fix: Migrate from `pg.members` or `pg.schema` to using Python type annotations (e.g., `field_name: pg.typing.Int()` or `field_name: int = pg.symbolic.Field(...)`) for declaring symbolic fields within `@pg.symbolize`d classes.

Install

pip install pyglove Install PyGlove

Imports

pyglove
```
import pyglove as pg
```

Quickstart

Define a symbolic class using `@pg.symbolize` and instantiate it. The object can then be used like a regular Python object.

import pyglove as pg

@pg.symbolize
class Hello:
  def __init__(self, subject):
    self._greeting = f'Hello, {subject}!'

  def greet(self):
    print(self._greeting)

hello = Hello('World')
hello.greet()

view raw JSON →