Skip to main content

List Hydra Sweeper plugin

Project description

List Sweeper plugin for Hydra

Sweeper plugin for Hydra which creates a list option additionally to the cartesian product ("grid"), which allows to sweep over the zipped list of parameters. This allows to test only a subset of the cartesian product and it is useful for small hyperparameter searches.

Installation

pip install hydra-list-sweeper

This will install the plugin in your current environment.

You can check if the plugin is installed by adding --info plugins to your command line. The plugin should be listed in the output as hydra_plugins.list_sweeper_plugin.list_sweeper.

In order to enable this plugin, you need to override the default sweeper in your configuration file:

defaults:
  - _self_
  # override this to use the new list sweeper:
  - override hydra/sweeper: list

Usage

List sweeper uses the a similar syntax as the standard sweeper, but instead of a params key, it uses grid_params and list_params:

hydra:
  mode: MULTIRUN
  sweeper:
    # standard grid search
    grid_params:
      env: 5_clubs_juggling, balancing_stick
    # additional list sweeper
    list_params:
      algorithm.lr: 0.001, 0.0001
      algorithm.beta_1: [0.9, 0.99]  # both notations work

This configuration will create 4 jobs:

env=5_clubs_juggling, algorithm.lr=0.001, algorithm.beta_1=0.9
env=5_clubs_juggling, algorithm.lr=0.0001, algorithm.beta_1=0.99
env=balancing_stick, algorithm.lr=0.001, algorithm.beta_1=0.9
env=balancing_stick, algorithm.lr=0.0001, algorithm.beta_1=0.99

Basically, it grids over all grid params, creating the standard cartesian product, and then for each of these combinations, it creates a job for each of the list params. You can additionally overwrite single values with command line arguments, and even define your grid_params in the command line:

hydra:
  mode: MULTIRUN
  sweeper:
    # additional list sweeper
    list_params:
      algorithm.lr: 0.001, 0.0001
      algorithm.beta_1: [0.9, 0.99]  # both notations work

Combined with this command

python my_app.py env=5_clubs_juggling,balancing_stick

will produce the same results as the first example. Also, you can override configs with the command line and the grid_params:

hydra:
  mode: MULTIRUN
  sweeper:
    # standard grid search
    grid_params:
      env: 5_clubs_juggling, balancing_stick
    # additional list sweeper
    list_params:
      algorithm.lr: 0.001, 0.0001
      algorithm.beta_1: [0.9, 0.99]  # both notations work

Combined with this command:

python my_app.py algorithm.epsilon=1.0e-4

will produce the same results as the first example, but epsilon will be set to 1.0e-4 for all jobs.

If you remove the list_params section, it will behave exactly as the standard grid sweeper (at least it should do, if you find a bug, please report it).

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

hydra-list-sweeper-1.0.1.tar.gz (5.9 kB view details)

Uploaded Source

Built Distribution

If you're not sure about the file name format, learn more about wheel file names.

hydra_list_sweeper-1.0.1-py3-none-any.whl (5.2 kB view details)

Uploaded Python 3

File details

Details for the file hydra-list-sweeper-1.0.1.tar.gz.

File metadata

  • Download URL: hydra-list-sweeper-1.0.1.tar.gz
  • Upload date:
  • Size: 5.9 kB
  • Tags: Source
  • Uploaded using Trusted Publishing? No
  • Uploaded via: twine/6.0.1 CPython/3.11.7

File hashes

Hashes for hydra-list-sweeper-1.0.1.tar.gz
Algorithm Hash digest
SHA256 56e760fd991ae8d7e5ba4663031cdc4aa240dfe970415508bbcf765279660a35
MD5 a5c080f49a51ea6f1bfff2d30a793204
BLAKE2b-256 49aebdb7942c61f0c2795242e6233de08111d8897c2180683b2da196eb5c2dd2

See more details on using hashes here.

File details

Details for the file hydra_list_sweeper-1.0.1-py3-none-any.whl.

File metadata

File hashes

Hashes for hydra_list_sweeper-1.0.1-py3-none-any.whl
Algorithm Hash digest
SHA256 3f172aa8da3e16e8056ccfad8af8377b212d9b2357f36ac416cd95f289949f71
MD5 0b4903409b40b4c83410fe56450a1a91
BLAKE2b-256 e580cc57d6d8552aaee96a87e35c0cbe098fd680e9144f8c589f0306ac2a4aa2

See more details on using hashes here.

Supported by

AWS Cloud computing and Security Sponsor Datadog Monitoring Depot Continuous Integration Fastly CDN Google Download Analytics Pingdom Monitoring Sentry Error logging StatusPage Status page