GNN API¶

Main components¶

TrainingConfig: training/task configuration.
GNNConditionEstimator: fit, evaluate, save/load, and predict interface.
train_gnn_condition_estimator: one-shot training helper.
SparseMatrixGNN: neural model implementation.
MatrixConditionDataset / MatrixGraph: graph dataset structures.

Training example¶

from sparse_kappa import TrainingConfig, train_gnn_condition_estimator

train_samples = [
    {'matrix': A, 'condition_number': 12.0, 'norm_Ainv': 0.6},
]

config = TrainingConfig(target='condition', norm=2, epochs=50, lr=1e-3)
estimator = train_gnn_condition_estimator(train_samples, config=config)

Prediction example¶

pred = estimator.predict(A)
print(pred)