Intelligent checkpointing framework for Python-based machine learning and scientific computing. Under development as part of a research project at the University of Illinois at Urbana-Champaign.

Installation

Run the following command in a virtual environment.

python setup.py install

Jupyter Integration

Run Jupyter after installing kishu. In your notebook, you can enable kishu with the following command.

Basic Usage

from kishu import init_kishu
init_kishu()

Then, all the cell executions are recorded, and the result of each cell execution is checkpointed.

Working with Kishu

init_kishu() adds a new variable _kishu (of type KishuJupyterExecHistory) to Jupyter's namespace. The special variable can be used for kishu-related operations, as follows.

Browse the execution log.

_kishu.log()

See the database file.

_kishu.checkpoint_file()

Restore a state.

_kishu.checkout(commit_id)

Checkpoint Backend

Deploy a restful server.

flask --app kishu/backend run

Deployment

The following command will upload this project to pypi (https://pypi.org/project/kishu/).

bash upload2pypi.sh