HyperPyYAML

1.2.3 · active · verified Sat Apr 11

HyperPyYAML is a Python library that extends the YAML syntax for enhanced interaction with Python objects and better hyperparameter definition. It is actively maintained, with a recent version 1.2.3, and typically releases updates as needed for bug fixes and new features.

Warnings

breaking HyperPyYAML allows arbitrary code execution through its custom tags, which is a feature for flexible configuration. Treat all YAML files loaded with HyperPyYAML as you would Python code. Loading untrusted YAML files can lead to security vulnerabilities.
Fix: Only load YAML files from trusted sources. Review any YAML configuration before using it with HyperPyYAML, especially files from external or untrusted origins.
breaking Compatibility issues with `ruamel.yaml` versions have occurred. Specifically, versions of `ruamel.yaml` 0.19.0 and above can cause breakage due to API changes.
Fix: Upgrade HyperPyYAML to version 1.2.3 or newer, which explicitly pins `ruamel.yaml<0.19.0` in its dependencies to ensure compatibility. If manually managing dependencies, ensure `ruamel.yaml` is below `0.19.0`.
gotcha The `!include` tag in older versions could break due to incorrect handling of overrides, particularly when no overrides were provided or the structure was unexpected.
Fix: Upgrade to HyperPyYAML 1.2.3 or higher, where this issue has been resolved.
gotcha The `!ref` tag creates a shallow reference to the original YAML node, meaning changes to the referenced object will affect all references. If a deep copy is desired, use the `!copy` tag instead.
Fix: Always use `!copy` when you need an independent instance of an object or value, ensuring that modifications to one do not unexpectedly alter others.
deprecated Older versions of HyperPyYAML had issues with Python 3.8+ compatibility, particularly concerning `ast.Constant`.
Fix: Upgrade to HyperPyYAML 1.2.3 or higher for full Python 3.8+ compatibility.

Install

pip install hyperpyyaml Install with pip

Imports

load_hyperpyyaml

from hyperpyyaml import load_hyperpyyaml

Quickstart

This quickstart demonstrates how to load a YAML string containing HyperPyYAML extensions. It shows the use of `!ref` for referencing other parameters and `!new` for instantiating Python objects (like PyTorch modules) directly from the YAML.

import torch
from hyperpyyaml import load_hyperpyyaml

example_hyperparams_yaml = """
base_channels: 32
kernel_size: 11
padding: !ref <kernel_size> // 2
layer1: !new:torch.nn.Conv1d
  in_channels: 1
  out_channels: !ref <base_channels>
  kernel_size: !ref <kernel_size>
  padding: !ref <padding>
model: !new:torch.nn.Sequential
  - !ref <layer1>
  - !new:torch.nn.LeakyReLU
"""

# Using load_hyperpyyaml to parse the YAML string
loaded_hparams = load_hyperpyyaml(example_hyperparams_yaml)

print(loaded_hparams['base_channels'])
print(loaded_hparams['padding'])
print(loaded_hparams['model'])

view raw JSON →