Fiddle: A Python-first configuration library

0.3.0 · active · verified Sun Apr 12

Fiddle is a Python-first configuration library designed particularly for machine learning applications. It enables deep configurability of parameters in a program, allowing configurations to be expressed in readable and maintainable Python code. It aims to reduce boilerplate and provides good error messages. The library is currently at version 0.3.0 and is actively developed by Google.

Warnings

breaking Fiddle is currently in 'Development Status :: 3 - Alpha'. This means the API is not yet stable, and breaking changes may occur in future releases.
Fix: Be prepared for API adjustments and refer to release notes when upgrading. Consider pinning minor versions in production environments.
gotcha Using mutable default arguments (e.g., lists, dictionaries) in functions configured by Fiddle can lead to unexpected shared state across different configurations or calls. Python evaluates default arguments once when the function is defined.
Fix: Always use `None` as a default argument, and then initialize the mutable object inside the function if `None` is passed. Example: `def func(arg=None): arg = arg or []`.
gotcha `fdl.Config` and `fdl.Partial` behave differently when `fdl.build()` is called. `fdl.Config` directly invokes the callable and returns its result, while `fdl.Partial` returns a `functools.partial` object that needs to be called separately to get the final result.
Fix: Understand the distinction: use `fdl.Config` when you want the immediate result of the callable's execution, and `fdl.Partial` when you want a configurable, pre-filled callable that can be invoked later with additional arguments.

Install

pip install fiddle Basic Installation
pip install fiddle[flags] With CLI Flag Support (absl-py)

Imports

fiddle
```
import fiddle as fdl
```
The official documentation and examples consistently use the `fdl` alias for `fiddle`.
Config
```
from fiddle import Config
```

Quickstart

This quickstart demonstrates how to define configurations for Python callables using `fdl.Config` and `fdl.Partial`, and then build them using `fdl.build()`. `fdl.Config` directly builds the callable with the specified arguments, while `fdl.Partial` returns a `functools.partial` object that can be called later.

import fiddle as fdl

def create_greeting(prefix: str, name: str = "World", punctuation: str = "!"):
    return f"{prefix} {name}{punctuation}"

# Create a Fiddle configuration for the function
cfg = fdl.Config(create_greeting, prefix="Hello", name="Fiddle")

# Build the configured function to get the result
result = fdl.build(cfg)
print(f"Original Fiddle build: {result}")

# Modify the configuration
cfg.punctuation = "!!"
result_modified = fdl.build(cfg)
print(f"Modified Fiddle build: {result_modified}")

# Demonstrate fdl.Partial, which returns a functools.partial object
partial_cfg = fdl.Partial(create_greeting, prefix="Hi")
partial_func = fdl.build(partial_cfg) # This returns a functools.partial object
result_partial = partial_func(name="Python") # Call the partial function
print(f"Partial Fiddle build: {result_partial}")

view raw JSON →