@haidark/experiments_ext v1.0.0

JupyterLab-Experiments

Adds "experiments" to JupyterLab notebooks that allow a user to share experimental design and results easily in a single notebook.

Experiments are custom cell execution orders defined by the user that show exactly how a result is generated in the notebook. The experiments are saved as part of the meta-data of the notebook and can be easily shared.

This extension adds a small set of UI elements to JupyterLab:

• A dropdown menu showing the available experiments
• Cell index labels

Although experiments are usually added after the code in a notebook is written, adding/removing/moving cells are smartly handled and the existing experiments are automatically updated to reflect changes to the notebook.

Basic Usage

• Click on the "Select Experiment" button and use the dropdown menu to select an experiment to run or add an experiment.
• Adding an experiment requires a label ("L2 Regularization") and a cell execution order ("4-5-6-9-12-13-14-8"). The cell execution order must be a list of valid cell indices separated by the dash (-) character.
• Click the "fast forward" icon to run the selected experiment.

Handling Changes

• Once a set of experiments are defined, changes to the order of the cells (add/remove/move) in the notebook will automatically update all experiments so that they point to the same cells.
• If all the cells that are part of an experiment are removed, the experiment will also be removed.
• To edit an existing experiment, enter the experiment label in the input field and assign it a new cell execution order.

Prerequisites

• JupyterLab

Installation

jupyter labextension install @haidark/experiments_ext

Development

For a development install (requires npm version 4 or later), do the following in the repository directory:

npm install
npm run build
jupyter labextension link .

To rebuild the package and the JupyterLab app:

npm run build
jupyter lab build

Authors

This extension was developed at the Rensselaer Polytechnic Institute Data Science Research Center by Haidar Khan Ph.D and Sadia Khan.