Serialization and Configuration Management

In addition to initializing environments using predefined identifiers, Sinergym supports saving and restoring complete environment configurations through the to_dict and from_dict methods. This functionality allows users to export, modify, and reload environment setups with ease, thereby promoting reproducibility and simplifying the sharing of experimental configurations.

config = env.to_dict()
# Save as YAML or reuse directly
new_env = EplusEnv.from_dict(config)

Although this method allows configuration persistence, serializing a Sinergym environment is not a trivial task. The environment may contain nested components, such as custom wrappers or references to complex objects, which are not easily expressed using basic data types.

To address this, Sinergym includes a dedicated serialization module (sinergym.utils.serialization) that leverages PyYAML to enable robust serialization and deserialization of both the environment and its associated wrappers. This module automates the transformation of complex Python objects into a YAML-compatible format, facilitating the saving and restoration of complete environment states.

Environment Configuration Serialization

When a Sinergym environment is instantiated, its configuration is automatically serialized and saved as a YAML file (env_config.pyyaml) within the environment’s designated output folder. This behavior ensures that every experiment includes a traceable and reproducible environment setup by default, even when the environment has been modified using user-specific configurations.

Wrapper Serialization and Restoration

Wrappers applied to an environment are not included in the default environment serialization and must be handled separately. Sinergym provides two utility functions for this purpose:

  • get_wrappers_info: Extracts metadata from all applied wrappers and serializes it.

  • apply_wrappers_info: Reconstructs and reapplies wrappers based on previously saved metadata.

Example: Serialization and Restoration

Wrapper configurations are not automatically serialized like the environment configuration. To save and restore wrapper configurations, you need to manually serialize the wrapper information and then apply it when loading the environment.

import gymnasium as gym
from sinergym.utils.common import get_wrappers_info, apply_wrappers_info
from sinergym.utils.wrappers import NormalizeObservation, NormalizeAction
from sinergym.envs.eplus_env import EplusEnv
import yaml

# Environment creation and wrapping
env = gym.make('Eplus-5zone-hot-continuous-stochastic-v1')
env = NormalizeObservation(env)
env = NormalizeAction(env)

# Serialize wrapper information
# By default, this will be saved in the environment's output directory
wrappers_info = get_wrappers_info(env) # Apply wrappers to the environment

By default, if no save path is specified, the wrapper configuration will be stored as wrappers_config.pyyaml in the environment’s output folder (accessible via env.get_wrapper_attr(‘workspace_path’)`).

To reload a complete environment, including its wrappers:

# Load environment parameters
with open('<env_output_folder>/env_config.pyyaml', 'r') as f:
    env_params = yaml.load(f, Loader=yaml.FullLoader)

# Modify environment parameters if needed...

# Restore environment
env = EplusEnv.from_dict(env_params)

# Restore wrappers (from file or dictionary)
env = apply_wrappers_info(env, '<env_output_folder>/wrappers_config.pyyaml')
# Alternatively:
env = apply_wrappers_info(env, wrappers_info)

This feature is used in Sinergym’s scripts in Usage.

Key Features

  • Automatic environment serialization on initialization (saved as env_config.pyyaml`).

  • Wrapper metadata extraction and persistence via get_wrappers_info.

  • Flexible restoration: Wrapper configurations can be reapplied using either a YAML file path or a Python dictionary.

  • Default paths: If no path is provided, serialization files are stored in the environment’s output folder.

  • Experiment reproducibility: Facilitates the sharing and exact reproduction of complex experimental setups.