FitTrace

class nyx.infer.convergence.FitTrace[source]

Bases: object

The path a fit took, sampled once per recorded solver step.

Variables:

  • steps (numpy.ndarray, shape (n_frames,)) – Solver step index of each frame; steps[0] is 0.

  • loss (numpy.ndarray, shape (n_frames,)) – Scalar loss at each frame, evaluated before that step was applied.

  • history (dict of str to numpy.ndarray) – One entry per probe, stacked over frames, each of shape (n_frames,) + probe_output_shape.

  • model (pytree) – The model at the last recorded frame.

Methods Summary

keys()

Names of the recorded probes.

Methods Documentation

keys()[source]

Names of the recorded probes.

Return type:

Iterator of str

__init__(steps, loss, history, model=None)[source]
classmethod __new__(*args, **kwargs)