Jikken is a lightweight cli experiment manager for scientific experiments written in python.
It makes very few assumptions on the experiment and requires almost zero code changes to the scripts. In fact the only assumption is that the main function of the script to be run accepts a positional argument with either a directory with json/yaml config files or a path to single json/yaml file. That's it. Optionally if the code of the experiment is in a git repo, the git info will be added as well.
- Python 3.{5,6} code
- Support for TinyDB, MongoDB, and ES
- tagging of experiments
- CLI to access experiment data
- only requires the script to load the variables from a file or folder
- support for json/yaml configs
- understands git directories
Jikken can be installed from pip. This installation is compatible with Linux/Mac OSX and Python 3.5 or greater is required. To install the package use:
or if the default python on your system is 2.7 use instead
pip3 install jikken
The main goal of Jikken is to require as little changes to your experiment code as possible. It's main assumption is that you have a script that runs an experiment and that the scripts accepts at least a positional argument with the location of the experiment's config. e.g. Let's assume you have an experiment runs by running my_experiment.py and it reads its configuration from a yaml file myconfig.yaml you would run it with:
python my_experiment.py myconfig.yaml
Then in order to let jikken record the experiment you would run instead:
jikken run my_experiment.py -c myconfig.yaml -n "my first experiment"
Jikken will then capture the config, Start the experiment, capture stdout and stderr, and save the information in the database as set in the config file.
The full documentation of jikken can be found here
- Fork it here
- Create your feature branch (git checkout -b feature/fooBar)
- Commit your changes (git commit -am 'Add some fooBar')
- Push to the branch (git push origin feature/fooBar)
- Create a new Pull Request
- 0.2.1
- Fixed bug with updated monitored (and added test)
- Added cli delete command for both experiments and multistage experiments
- refactored documentation a bit
- 0.2.0
- Added mongodb and ES support
- Added support for multistage experiments
- started writing documentation on read the docs
- 0.1.0
- Work in progress
We use SemVer for versioning. For the versions available, see the tags on this repository
- Agis Oikonomou - Initial work
This project is licensed under the MIT License - see the LICENSE file for details
- Brian Okken and his great book on python testing and a great influence on the structure of the code: Python Testing with Pytest.
- Francois Chollet and his book Deep Learning with Python. The examples of jikken are all based on the ones from the book.