qbiocode.apps.quvine.baselines.hyperparameter_loader module

Hyperparameter Loader for Tuned Methods

This module loads pre-tuned hyperparameters from JSON files and overrides config defaults. Each dataset has its own tuning file with task-specific parameters (node_classification, link_prediction, node_ranking).

Summary

Classes:

HyperparameterLoader

Loads and manages tuned hyperparameters from JSON files.

Functions:

get_global_loader	Get the global hyperparameter loader.
load_and_override_config	Convenience function to override config with tuned hyperparameters.
set_global_loader	Set the global hyperparameter loader.

Reference

class HyperparameterLoader(tuning_dir=None, dataset_name=None)

Bases: object

Loads and manages tuned hyperparameters from JSON files.

Example JSON structure:

{
    "node2vec": {
        "node2vec": {
            "node_classification": {"best_params": {...}, "best_score": 0.5},
            "link_prediction": {"best_params": {...}, "best_score": 0.9},
            "node_ranking": {"best_params": {...}, "best_score": 0.4}
        }
    },
    ...
}

__init__(tuning_dir=None, dataset_name=None)

Initialize hyperparameter loader.

Parameters:

• tuning_dir (Optional[str]) – Directory containing tuning JSON files

• dataset_name (Optional[str]) – Name of the dataset (e.g., ‘scale_free’, ‘BioPlex3_autism’)

load_hyperparameters()

Load hyperparameters from JSON file.

Tries multiple filename patterns: 1. {dataset_name}_tuning_by_task.json (new format) 2. {dataset_name}_aggregated.json (legacy format)

Return type:

bool

Returns:

True if loaded successfully, False otherwise

get_method_params(method_name, task='node_classification')

Get tuned parameters for a method and task.

Parameters:

• method_name (str) – Name of the method (e.g., ‘node2vec’, ‘gat_ctqw_heat’)

• task (str) – Task type (‘node_classification’, ‘link_prediction’, ‘node_ranking’)

Return type:

Optional[Dict[str, Any]]

Returns:

Dictionary of best parameters, or None if not found

override_config(config, method_name, task='node_classification')

Override config object with tuned hyperparameters.

Parameters:

• config – Config dataclass object

• method_name (str) – Name of the method

• task (str) – Task type

Returns:

Updated config object (new instance via dataclass replace)

override_nested_config(config, method_name, task='node_classification')

Override nested config objects (for GAT/GraphGPS with model and train configs).

Parameters:

• config – Config dataclass with nested model and train configs

• method_name (str) – Name of the method

• task (str) – Task type

Returns:

Updated config object

set_global_loader(loader)

Set the global hyperparameter loader.

get_global_loader()

Get the global hyperparameter loader.

Return type:

Optional[HyperparameterLoader]

load_and_override_config(config, method_name, task='node_classification')

Convenience function to override config with tuned hyperparameters.

Parameters:

• config – Config dataclass object

• method_name (str) – Name of the method

• task (str) – Task type

Returns:

Updated config object