shephex 0.2.2

Computational experiment manager. Execute parameterized Python function on Slurm directly from a python script

shephex

A Python package for managing and executing computational experiments.

Features

• Easy management of parameter sweeps
• Execution of python function or scripts with input sets.
• Simple SLURM Scheduling from Python.
• Extendable execution class to adapt to other job management systems.

The example show how easy it is to submit an arbitrary Python function for execution on Slurm. The function in this case is a biased-random walk in 2D for which 50 different parameterizations are submitted and executed.

import numpy as np
import shephex

@shephex.hexperiment()
def biased_random_walk_2d(steps: int, bias_x: float, bias_y: float) -> tuple:
    """
    Simulate a 2D biased random walk and return the final position.
    """    
    # Movement choices: (-1, 1) for left/right (x) and down/up (y)
    moves_x = np.random.choice([-1, 1], size=steps, p=[1 - bias_x, bias_x])
    moves_y = np.random.choice([-1, 1], size=steps, p=[1 - bias_y, bias_y])
    
    final_x = np.sum(moves_x)
    final_y = np.sum(moves_y)
    
    return final_x, final_y  # Return final (x, y) position

if __name__ == '__main__':

    experiments = []
    for steps in [10, 100, 250, 500, 1000]:
        for bias_x in [0.5, 0.6]:
            for repeat in range(5):
                experiments += [biased_random_walk_2d(steps=steps, 
                                                    bias_x=bias_x, 
                                                    bias_y=0.5)]

    executor = shephex.SlurmExecutor(ntasks=1, partition='q48', time='00:10:00')
    executor.execute(experiments)