Strategery

Installation

This package has zero dependencies, so don't be shy.

pip install strategery

About

Strategery is an approach to structuring code, built around feature extraction tasks in machine learning. It enables flexibility and configurability for complex, interdependent graphs of functions. I'm no expert in design patterns, but I think it's somewhere between chain-of-responsibility and strategy.

You're trying to extract some features, which are the output of numerous functions. You find yourself reusing the same algorithms in a bunch of scripts, but sometimes only some of them, and sometimes in a different order, and sometimes you want to discard interstitial data. Sometimes you perform an expensive calculation in the middle, but you want to share that result between other functions. Strategery might be for you.

For a toy example, let's say the features you're extracting are a Minimum, Mean, and AbsoluteValueMinimum.

from strategery import fed_by

@fed_by('ParamListOfNumbers')
def Minimum(numbers):
    ''' The least '''
    return min(numbers)


@fed_by('ParamListOfNumbers')
def Mean(numbers):
    ''' AKA Average '''
    return 1.0 * sum(numbers) / len(numbers)


@fed_by(Minimum)
def AbsoluteValueMinimum(minimum):
    ''' A highly contrived task for example purposes '''
    return abs(minimum)

Hold up, @fed_by()?

OK, so @fed_by() is a decorator that will mark up your tasks so it can tell the engine what its own dependencies are. Then the engine can calculate a strategery for how to get to the results you requested, and if you provided all the necessary components, it resolve all the dependencies and hand off values between your functions. @fed_by accepts a reference to a function (if you need the results from another task), or a string (if you want to provide the input as a parameter).

All right, that's it for setup. Let's crunch some numbers. When you're calling strategery.execute, the targets parameter is a list of the results you want, and the preprocessed parameter lets you provide information for the engine to start with.

import strategery
import sys

# Enable verbose logging:
strategery.engine.logger = sys.stderr

# Provide a list of numbers, get AbsoluteValueMinimum, let strategery figure out the rest.
mean, absolute_value_minimum = strategery.execute(
        targets=[
            Mean,
            AbsoluteValueMinimum
        ],
        input={
            'ParamListOfNumbers': [-1, -2, -3, -4, -5]
        }
    )

# Disable verbose logging:
strategery.engine.logger = strategery.logging.BypassLogger()

print("")
print('Input: [-1, -2, -3, -4, -5]')
print(f'Mean: {mean}')
print(f'AbsoluteValueMinimum: {absolute_value_minimum}')