logrun

A package for experiment logging in Python.

Features include:

• Stores relevant source code (only what is necessary) with the experiments automatically;
• Easily fetch the experiments that generated a given file;
• Easily fetch the experiments that were generated by a given script;
• Shouldn't use tons of storage;
• Everything runs locally -- no possibly-sketchy third-party will be getting access to your code and experiments.

Install

logrun can be installed via pip:

pip install logrun

Example usage

The following example trains a model and registers an experiment that outputted the file trained_model.pth:

from logrun.utils.general import outpath
import pickle


...  # your experiment, yada yada yada
model = ...
...  # your experiment, yada yada yada

with open(outpath("trained_model.pickle"), 'wb') as file:
    pickle.dump(model, file)

Note that you only needed to import the module and use the outpath function here. Everything else is neatly encapsulated in the logrun.internals module.

Documentation

The documentation is available at https://dccsillag.github.io/logrun/logrun/.

To generate it manually, use pdoc3:

pdoc --html logrun -o docs