Skip to content

Command Line Interface

Different parts of AutoRA experiments can require very different computational resources. For instance:

  • The theorist and experimentalist might require training or use of neural networks, and benefit from high performance computing (HPC) resources for short bursts – minutes or hours.
  • The experiment runner might post an experiment using a service like "Prolific" and poll every few minutes for hours, days or week until the experimental data are gathered.

Running the experiment runner with the same resources as the theorist and experimentalist in this case would be wasteful, and may be prohibitively expensive.

To solve this problem, AutoRA comes with a command line interface (CLI). This can be used with HPC schedulers like SLURM to run different steps in the cycle with different resources.

You can use the CLI if the following conditions are true:

  1. Every part of s can be successfully pickled.
  2. You can write each step of your experiment as a single importable function which operates on a state and returns a state: 
    from example.lib import initial_state, experimentalist, experiment_runner, theorist
s = initial_state()
for i in range(3):
    s = experimentalist(s)
    s = experiment_runner(s)
    s = theorist(s)

Contents of this section: