FitResult

Contents

FitResult#

class braintools.fit.FitResult(best_params, best_loss, params, losses, history, converged, backend, raw=None)#

Outcome of Fitter.fit().

Variables:

  • best_params (dict) – {name: value} at the best start, in the constrained (physical) space, with units restored.

  • best_loss (float) – Loss of best_params.

  • params (dict) – {name: values} of every start (leading axis S), constrained, with units. For 'nevergrad' these are the candidates of the last generation.

  • losses (jax.Array) – Final loss of every start, shape (S,).

  • history (jax.Array) – Losses per optimisation step, shape (S, n_steps). For 'grad' the loss after every iteration (entries after a start stopped repeat its final loss); for 'scipy' the loss after every SciPy iteration, padded the same way; for 'nevergrad' the loss of candidate s of every generation.

  • converged (jax.Array) – Boolean (S,). 'grad': the start met the convergence test; 'scipy': OptimizeResult.success; 'nevergrad': the candidate’s loss is finite.

  • backend (str) – 'grad', 'scipy' or 'nevergrad'.

  • raw (Any) – Backend output: the MultiStartResult, the list of SciPy OptimizeResult objects, or the NevergradOptimizer.

Contents