Argparse, easy reloading of command line arguments

command-line-parser, hyperparameters, python
pip install argload==0.1.1


Argload, easy reloading of command line arguments

Argload is a python package facilitating the reloading of command line arguments.

As an example is worth a thousand words, let's explain the use cases of argload with an example bit of code.

Say you have a bunch of experiments to run depending on two hyperparameters, a and b. You define your parser with argparse as follow:

import argparse

parser = argparse.ArgumentParser()
parser.add_argument('--a', type=float, default=1e-3)
parser.add_argument('--b', type=float, default=2.)
args = parser.parse_args()

Now say you have 2 very long experiments to run, one with a=1e-3, b=3, and the other one with a=1e-4, b=1. If both experiments were very short, everything would be fine, you would launch each of them independantly, and even if they crash, simply restarting them using your bash history is both simple and fast.

But if they are very long, there is a non negligible chance for your command line not to be in your bash history anymore, or for you to have forgotten your exact hyperparameters (OK, two is easy, 32 less so, 32 hyperparameters on 54 experiments even less so...).

That's where argload can come in handy. Now simply wrap your parser in an ArgumentLoader instance in the following way:

import argparse
import argload

parser = argparse.ArgumentParser()
parser.add_argument('--a', type=float, default=1e-3)
parser.add_argument('--b', type=float, default=2.)
parser = argload.ArgumentLoader(parser, to_reload=['a', 'b'])
args = parser.parse_args()

Your script takes one additional required argument logdir, the directory where your current argument will be stored (and typically the directory where all your results will be stored), and two optional flags overwrite and dump. Say you want to store the results of your first experiment in the exp1. After creating this directory, the first time you run your script, with the logdir argument properly set,

python --logdir exp1 --a 1e-3 --b 3

argload stores your arguments in the exp1 directory. Now, the next time you run your script in exp1,

python --logdir exp1

your hyperparameters will be automatically loaded.

I you specify the overwrite flag, you can overwrite previous values of your hyperparameters. For instance, running

python --logdir exp1 --overwrite --a 1e-2

will relaunch with arguments a=1e-2, b=3. However, the new value for a will not be stored in your directory, if you relaunch

python --logdir exp1

the values you will be using are still a=1e-3, b=3.

If you want to store your new setting inside of your experiment directory, issue the command

python --logdir exp1 --overwrite --dump --a 1e-2

The next time you will launch your script in exp1, the values used will be a=1e-2, b=3.