Skip to content

A set of hackable helpers for python-only (deep learning) experiment scripts

Notifications You must be signed in to change notification settings

wheatdog/EZExps

Repository files navigation

EZExps

A set of hackable helpers for python-only (deep learning) experiment scripts

Features

  • Track local dependencies of your experiment script. For reproducibility, it won't run your experiment until those changes are committed via git.
  • Store experiment information into MongoDB after finishing.
  • Upload experiment information to Google Sheets if you have OAuth2 credentials for Google API.

Tracked experiment information

  • Arguments
  • Artifacts
  • Purpose
  • Source files
  • Start/End/Elapsed Time
  • Stdout

Prerequisites

You will need

There are some assumptions for the .py we can track.

  • Must have a function named get_args, which returns an object like ArgumentParser.parse_args() returns.
  • Must have a function named main, which take the output of get_args as the only argument and return the experiment artifacts.

You can check out example for more detail.

Installation

pip install -r requirements.txt

Setup ezexps.ini for your project, check out ezexps.ini.

Helper scripts

  • exps.py
  • gsheet_uploader.py
  • local_dependency.py

Example

cd example
python ../exps.py 'lr=0.1 test' main.py --lr 0.1 --epochs 1

Todo

  • A task queuer: will check if gpu is available and assign experiments
  • A daemon validation tool

About

A set of hackable helpers for python-only (deep learning) experiment scripts

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages