pip install -U inex-launcherinex-launcher lets you describe an entire experiment in a single YAML file. The plugins list defines the initialization order of handlers. For each handler you only need to provide the module path and its options. The execute section describes the final method that will be called.
plugins:
- loader
- model
- trainer
loader:
module: myproject.data/Loader
options:
path: data/train.csv
model:
module: myproject.nn/Model
options:
hidden: 128
trainer:
module: myproject.training/Trainer
imports:
model: plugins.model
data: plugins.loader
options:
epochs: 10
execute:
method: myproject.training/run
imports:
trainer: plugins.trainerRun this configuration with:
inex config.yamlYou can fine-tune the experiment by merging it with another file using --merge and overriding specific values with --update:
inex config.yaml -m experiment.yaml -u trainer.epochs=20 model.hidden=64This approach lets you modify parameters or even replace handlers with just a few lines of YAML—no need to edit high‑level Python files.
Key benefits of this approach:
- High modularity – modules can be swapped directly in the configuration thanks to the plugin-based structure.
- Clean structure – a clear sequence of
pluginsfollowed by a singleexecuteblock keeps the file readable. - Flexibility – YAML allows complex nested structures, so even advanced scenarios fit in one file.