kerighan/search

A simple class for brute force parameter search

Search

Overview

This a simple Python class that allows to concurrently minimize or maximize arbitrary functions, by brute forcing a grid of parameters.

Disclaimer

This implementation has only two aims: ease and speed of use, not algorithm efficiency.

Dependencies

• numpy
• tqdm

Tutorial

Simply instantiate a Search object, define the parameters space with the method grid(**kwargs) and use the method minimize(function) to minimize a given function.

from search import Search
import numpy as np


# given data
x = np.linspace(0, 10, 100)
y = 3 * x + 2  # 3 and 2 are the parameters to infer


# loss function
def loss(x, y, a, b):
    y_pred = x * a + b
    loss = np.mean((y - y_pred)**2)
    return loss


search = Search(workers=4)  # 4 workers Pool
# `grid` requires iterable parameters,
# so giving [x] acts as one parameters `x`
search.grid(
    a=np.linspace(0, 11, 100),
    b=np.linspace(0, 11, 100),
    x=[x],
    y=[y]
)
data = search.minimize(loss)
print(data[0]["parameters"])  # a = 3, b = 2