mixle.experimental.model_economy module

P14 (experimental, speculative) – model economies: trading verified components beats isolation.

Two agents hold complementary training data. A buyer may adopt coefficients offered by the other agent, but component selection is performed on a buyer-local validation split and every reported MSE is measured later on an untouched final test split. The full-data-sharing oracle fits the exact union of the two agents’ training rows; it does not receive a separately generated, easier design.

Every train/selection/test matrix is sampled from the declared iid_standard_normal_scaled design distribution before the non-domain columns of training rows are masked. This construction works for underdetermined and overdetermined row/feature configurations and keeps isolation, trade, and oracle candidates on one data-generating process.

class EconomyReport(isolation_mse, trade_mse, oracle_mse, oracle_gain, trade_gain, recovered_fraction, adopted, selection_isolation_mse, selection_trade_mse, n_selection, n_test, selection_data_digest, test_data_digest, design_distribution='iid_standard_normal_scaled')[source]

Bases: object

Final-test report plus the separate selection receipt.

Parameters:

  • isolation_mse (float)

  • trade_mse (float)

  • oracle_mse (float)

  • oracle_gain (float)

  • trade_gain (float)

  • recovered_fraction (float)

  • adopted (int)

  • selection_isolation_mse (float)

  • selection_trade_mse (float)

  • n_selection (int)

  • n_test (int)

  • selection_data_digest (str)

  • test_data_digest (str)

  • design_distribution (str)

run_economy(*, n_features=10, cols_a=(0, 1, 2), cols_b=(5, 6, 7), n_train=400, n_selection=None, n_test=400, noise=0.1, seed=0)[source]

Measure isolation, validation-selected component trade, and exact training-data sharing.

Parameters:
Return type:

EconomyReport