Customize Hydra’s Configuration

Hydra is highly configurable. The way that Hydra runs jobs, logs information, manages job directories are among its many configurable components [1]. This How-To guide demonstrates how to configure Hydra from within a Python module [2].

To configure Hydra using hydra-zen we must update the default instance of hydra.conf.HydraConf (which is a dataclasses.dataclass()). Thus we can simply create our own hydra.conf.HydraConf instance and overwrite store’s entry; ZenStore will automatically place this under the appropriate group="hydra", name="config" entry in Hydra’s config store.

In this example, we configure Hydra to automatically change the runtime working directory to that of the job’s output directory. I.e., we set hydra.job.chdir=True:

Contents of my_app.py

from hydra_zen import zen, store
from hydra.conf import HydraConf, JobConf

def task():  # just an example task function - no important details
    import os
    print(f"working dir: {os.getcwd()}")

if __name__ == "__main__":
    store(HydraConf(job=JobConf(chdir=True)))  # store `hydra.job.chdir=True`
    store(task)  # create & store config for our task fn
    store.add_to_hydra_store()

    zen(task).hydra_main(config_path=None, config_name="task", version_base="1.2")

Let’s confirm that we have successfully changed Hydra’s configuration. By default, our app’s working dir should be job’s the time-stamped output directory.

Running our application using the default config.

$ pwd
foo/

$ python my_app.py
working dir: foo/outputs/2023-01-20/17-48-19

$ python my_app.py hydra.job.chdir=False
working dir: foo

Footnotes

[1]

See this documentation to learn about all of the ways that Hydra can be configured.

[2]

Hydra’s documentation for configuring Hydra focuses on yaml-based and CLI-based approaches. In contrast, hydra-zen promotes a pure-Python approach to Hydra and thus this guide follows suite.