Skip to content

Contribute An Experimentalist

AutoRA experimentalists are meant to return novel experimental conditions based on prior experimental conditions, prior observations, and/or prior models. Such conditions may serve as a basis for new, informative experiments conducted by an experiment runner. Experimentalists are generally implemented as functions that can be integrated into an Experimentalist Pipeline.

Experimentalist Module

Repository Setup

We recommend using the cookiecutter template to set up a repository for your experimentalist. Alternatively, you can use the unguided template. If you choose the cookiecutter template, you can set up your repository using


Make sure to select the experimentalist option when prompted. You can skip all other prompts pertaining to other modules (e.g., experiment runners) by pressing enter.


For an experimentalist, you should implement a function that returns a set of experimental conditions. This set may be a numpy array, iterator variable or other data format.


We generally recommend using 2-dimensional numpy arrays as outputs in which each row represents a set of experimental conditions. The columns of the array correspond to the independent variables.

Once you've created your repository, you can implement your experimentalist by editing the file in src/autora/experimentalist/name_of_your_experimentalist/. You may also add additional files to this directory if needed. It is important that the file contains a function called name_of_your_experimentalist which returns a set of experimental conditions (e.g., as a numpy array).

The following example illustrates the implementation of a simple experimentalist that uniformly samples without replacement from a pool of candidate conditions.

Example Experimentalist

import random
from typing import Iterable, Sequence, Union

random_sample(conditions: Union[Iterable, Sequence], n: int = 1):
    Uniform random sampling without replacement from a pool of conditions.
        conditions: Pool of conditions
        n: number of samples to collect

    Returns: Sampled pool


    if isinstance(conditions, Iterable):
        conditions = list(conditions)
    samples = conditions[0:n]

    return samples

Next Steps: Testing, Documentation, Publishing

For more information on how to test, document, and publish your experimentalist, please refer to the general guideline for module contributions .