OptimHistory.position_df()

OptimHistory.position_df(subset=None)

Converts the saved parameter position history into a data frame.

Parameters:

subset (Sequence[str] | None, default: None) – Optional sequence of parameter names to keep.

Raises:

TypeError – If position history was not saved.

Return type:

DataFrame

Examples

>>> import jax.numpy as jnp
>>> from liesel.optim import OptimHistory
>>> from liesel.optim.types import Position
>>> history = OptimHistory.from_epochs(
...     2, Position({"theta": jnp.array([1.0, 2.0])})
... )
>>> history.position_df().columns.tolist()
['epoch', 'theta0', 'theta1']
>>> no_position = OptimHistory.from_epochs(2, position=None)
>>> try:
...     no_position.position_df()
... except TypeError as error:
...     print("not saved" in str(error))
True