cmdline-brute 0.3

# Brute: easy grid search

## Introduction

This command-line script lets you run any command with many possible inputs in a distributed environment such as SGE, SLURM, or a local machine with many processors. Basic checkpointing is provided via the doit library, so that failed jobs might be re-run without re-running all jobs.

## Installation

Run:

python setup.py install

This should place the command brute in your path.

## Usage

Sample usage:

brute worker.py –foo 1,2,3 –bar x

will execute 3 tasks:

• worker.py –foo 1 –bar x

• worker.py –foo 2 –bar x

• worker.py –foo 3 –bar x

Each task will have a subdirectory where a run script and a log file will be created. The location where the subdirectories are created may be controlled via the –brute-dir flag.

Grid-specific options go in a configuration file. See examples/brute.conf for an example.

## Summarizing results

In addition to the brute command-line script, the scrape command is provided to facilitate summarizing the results of large grid jobs. This command has two main uses:

1. Obtain the job return status information. This is the default behavior, and produces a summary of the return status. Alternatively, –status-verbose prints the return status for all jobs instead of a summary.

2. For jobs which produce a single number as a result, scrape may be used to sort and summarize the jobs according to this score.

This second use is especially useful in machine learning applications. For this purpose, a Python script must be provided via –scraper. This script must provide the following method:

def scrape(PATH):

# TODO: implement code to scrape job output return SCORE

The result will be the jobs and their arguments, sorted by their scores. The –max argument may be used to limit the number of displayed results.