xnn.common.config.schema

Typed configuration schema.

A single nested dataclass tree is the one internal representation. Every frontend (YAML, argparse, Hydra) is just a different loader that produces this same Config. That is what makes the formats interchangeable – see loaders.py.

Classes

Config([model, data, optim, device, seed, ...])	Top-level experiment configuration.
DataConfig([train_path, val_path, ...])	Dataset paths and data-loading options.
ModelConfig([name, cutoff, n_features, ...])	Model architecture and hyperparameters.
OptimConfig([lr, weight_decay, epochs, ...])	Optimizer, schedule, and loss-weighting settings.