Home

Overview

hya is a library of custom OmegaConf resolvers designed to be used with Hydra. It provides a comprehensive set of resolvers for mathematical operations, path manipulation, data type conversions, and more.

Quick Start

The default resolvers are automatically registered when you import hya:

import hya
from omegaconf import OmegaConf

# Use resolvers in your configuration
conf = OmegaConf.create(
    {
        "batch_size": 32,
        "num_samples": 1000,
        "num_batches": "${hya.ceildiv:${num_samples},${batch_size}}",
        "learning_rate": "${hya.pow:10,-3}",  # 0.001
    }
)

print(conf.num_batches)  # Output: 32
print(conf.learning_rate)  # Output: 0.001

Using with Hydra

Create a configuration file config.yaml:

model:
  input_dim: 784
  hidden_dim: 256
  output_dim: 10

training:
  batch_size: 32
  num_epochs: 10
  learning_rate: ${hya.pow:10,-3}  # 0.001

paths:
  data_dir: ${hya.path:/data}
  model_dir: ${hya.iter_join:[${paths.data_dir},models,v1],/}

experiment:
  name: mnist_classifier
  id: ${hya.sha256:${experiment.name}}

Then use it in your Python script:

import hya
import hydra
from omegaconf import DictConfig


@hydra.main(version_base=None, config_path=".", config_name="config")
def main(cfg: DictConfig):
    print(f"Learning rate: {cfg.training.learning_rate}")
    print(f"Model dir: {cfg.paths.model_dir}")
    print(f"Experiment ID: {cfg.experiment.id}")


if __name__ == "__main__":
    main()

Registering Custom Resolvers

You can extend hya with your own custom resolvers:

from hya import get_default_registry

registry = get_default_registry()


@registry.register("multiply")
def multiply_resolver(x, y):
    return x * y


# Register the custom resolver with OmegaConf
registry.register_resolvers()

# Now use it in your configuration
conf = OmegaConf.create({"result": "${multiply:5,3}"})
print(conf.result)  # Output: 15

Available Resolvers

hya provides over 20 built-in resolvers organized into categories:

• Mathematical operations: add, sub, mul, truediv, floordiv, ceildiv, pow, sqrt, neg, max, min, exp, log, log10, sinh, asinh
• Constants: pi
• Path utilities: path, to_path, iter_join
• Utilities: len, sha256
• Optional (with dependencies):
• NumPy: np.array
• PyTorch: torch.tensor, torch.dtype
• Braceexpand: braceexpand

See the Resolvers page for complete documentation of all resolvers.

API stability

While hya is in development stage, no API is guaranteed to be stable from one release to the next. In fact, it is very likely that the API will change multiple times before a stable 1.0.0 release. In practice, this means that upgrading hya to a new version will possibly break any code that was using the old version of hya.

License

hya is licensed under BSD 3-Clause "New" or "Revised" license available in LICENSE file.