alkymi 0.0.7

alkymi - Pythonic task automation

alkymi ⚗️

Alkymi is a pure Python (3.6+) library for describing and executing tasks and pipelines with built-in caching and conditional evaluation based on checksums.

Alkymi is easy to install, simple to use, and has no dependencies outside of Python's standard library. The code is cross-platform, and allows you to write your pipelines once and deploy to multiple operating systems (tested on Linux, Windows and Mac).

Documentation, including a quickstart guide, is provided here.

Features

• Easily define complex data pipelines as decorated Python functions
  • This allows you to run linting, type checking, etc. on your data pipelines
• Return values are automatically cached to disk, regardless of type
• Efficiently checks if pipeline is up-to-date
  • Checks if external files have changed, bound functions have changed or if pipeline dependencies have changed
• No domain specific language (DSL) or CLI tool, just regular Python
  • Supports caching and conditional evaluation in Jupyter Notebooks
• Cross-platform - works on Linux, Windows and Mac
• Expose recipes as a command-line interface (CLI) using alkymi's Lab type

Sample Usage

For examples of how to use alkymi, see the quickstart guide.

Example code:

import numpy as np
import alkymi as alk

@alk.recipe()
def long_running_task() -> np.ndarray:
    # Perform expensive computation here ...
    hard_to_compute_result = np.array([42])
    # Return value will be automatically cached to disk
    return hard_to_compute_result

result = long_running_task.brew()  # == np.ndarray([42])

Or one of the examples, e.g. MNIST.

Upcoming Features

The following features are being considered for future implementation:

• Type annotations propagated from bound functions to recipes
• Support for call/type checking all recipes (e.g. by adding a check command to Lab)
• Cache maintenance functionality
• Parallel execution of pipeline

Known Issues

• alkymi currently doesn't check custom objects for altered external files when computing cleanliness (e.g. MyClass has a self._some_path that points to a file somewhere outside alkymi's internal cache)
• alk.foreach() currently only supports enumerable inputs of type List or Dict
• Recipes marked transient will always be dirty, and thus always require reevaluation. This functionality should be replaced by a proper means of creating recipes that don't cache outputs, but only run when needed to provide inputs for downstream recipes

Installation

Install via pip:

pip install --user alkymi

Or see the Installation page.

Testing

After installing, you can run the test suite (use the lint, coverage and type_check recipes to perform those actions):

python3 labfile.py brew test

License

alkymi is licensed under The MIT License as found in the LICENSE.md file