bencher.optuna_conversions

Attributes

`logger`
`_EXPECTED_PLOT_FAILURES`

Functions

`optuna_grid_search`(→ optuna.Study)	use optuna to perform a grid search
`param_importance`(→ panel.Column)
`summarise_trial`(→ list[str])	Given a trial produce a string summary of the best results
`sweep_var_to_optuna_dist`(...)	Convert a sweep var to an optuna distribution
`sweep_var_to_suggest`(→ object)	Converts from a sweep var to an optuna
`cfg_from_optuna_trial`(...)	Build a worker config from an optuna trial's suggested values.
`_plot_failure_pane`(plot_fn, exc, *[, unexpected])	Build the error pane that replaces a plot that failed to render.
`_append_safe`(row, plot_fn, args, *kwargs)	Append a plot to row, surfacing any exception instead of propagating.
`_append_safe_sized`(row, plot_fn, width, args, *kwargs)	Like _append_safe but sets a consistent width on the resulting plotly figure.

Module Contents

bencher.optuna_conversions.logger

bencher.optuna_conversions.optuna_grid_search(bench_cfg: bencher.bench_cfg.BenchCfg, trial_vars: list | None = None) → optuna.Study

use optuna to perform a grid search

Parameters:

bench_cfg (BenchCfg) – setting for grid search
trial_vars (list | None) – If provided, use these variables for the search space. Otherwise, filter bench_cfg.all_vars by optimize=True.

Returns:

results of grid search

Return type:

optuna.Study

bencher.optuna_conversions.param_importance(bench_cfg: bencher.bench_cfg.BenchCfg, study: optuna.Study, plot_width: int | None = None) → panel.Column

bencher.optuna_conversions.summarise_trial(trial: optuna.trial, bench_cfg: bencher.bench_cfg.BenchCfg) → list[str]

Given a trial produce a string summary of the best results

Parameters:

trial (optuna.trial) – trial to summarise
bench_cfg (BenchCfg) – info about the trial

Returns:

Summary of trial

Return type:

list[str]

bencher.optuna_conversions.sweep_var_to_optuna_dist(var: param.Parameter) → optuna.distributions.BaseDistribution

Convert a sweep var to an optuna distribution

Parameters:: var (param.Parameter) – A sweep var
Raises:: ValueError – Unsupported var type
Returns:: Optuna representation of a sweep var
Return type:: optuna.distributions.BaseDistribution

bencher.optuna_conversions.sweep_var_to_suggest(iv: bencher.variables.parametrised_sweep.ParametrizedSweep, trial: optuna.trial) → object

Converts from a sweep var to an optuna

Parameters:

iv (ParametrizedSweep) – A parametrized sweep input variable
trial (optuna.trial) – Optuna trial used to define the sample

Raises:

ValueError – Unsupported var type

Returns:

A sampled variable (can be any type)

Return type:

Any

bencher.optuna_conversions.cfg_from_optuna_trial(trial: optuna.trial, bench_cfg: bencher.bench_cfg.BenchCfg, cfg_type: type[bencher.variables.parametrised_sweep.ParametrizedSweep]) → bencher.variables.parametrised_sweep.ParametrizedSweep

Build a worker config from an optuna trial’s suggested values.

cfg_type is the worker class, not an instance – it is called here to construct one. Annotating it ParametrizedSweep makes a checker resolve cfg_type() through Parameterized.__call__ and infer a dict.

bencher.optuna_conversions._EXPECTED_PLOT_FAILURES

bencher.optuna_conversions._plot_failure_pane(plot_fn, exc: Exception, *, unexpected: bool = False)

Build the error pane that replaces a plot that failed to render.

unexpected marks an exception outside _EXPECTED_PLOT_FAILURES; it gets an extra ERROR log naming the type, because an unanticipated type here is a signal that the expected-failure tuple has fallen out of date.

bencher.optuna_conversions._append_safe(row, plot_fn, *args, **kwargs)

Append a plot to row, surfacing any exception instead of propagating.

Nothing is ever re-raised: this runs while the report is being built, after the sweep has already paid for every sample, so a crash here would throw away the whole run’s results to report one missing plot. A visible error pane costs nothing and keeps the rest of the report.

bencher.optuna_conversions._append_safe_sized(row, plot_fn, width, *args, **kwargs): Like _append_safe but sets a consistent width on the resulting plotly figure.