zfit.minimizers.baseminimizer module

Definition of minimizers, wrappers etc.

exception zfit.minimizers.baseminimizer.FailMinimizeNaN

Bases: Exception

args

with_traceback()

Exception.with_traceback(tb) – set self.__traceback__ to tb and return self.

class zfit.minimizers.baseminimizer.ZfitStrategy

Bases: abc.ABC

abstract minimize_nan(loss, params, minimizer, values=None)

Return type

float

class zfit.minimizers.baseminimizer.BaseStrategy

Bases: zfit.minimizers.baseminimizer.ZfitStrategy

minimize_nan(loss, params, minimizer, values=None)

Return type

float

class zfit.minimizers.baseminimizer.ToyStrategyFail

Bases: zfit.minimizers.baseminimizer.BaseStrategy

minimize_nan(loss, params, minimizer, values=None)

Return type

float

class zfit.minimizers.baseminimizer.PushbackStrategy(nan_penalty=100, nan_tolerance=30, **kwargs)

Bases: zfit.minimizers.baseminimizer.BaseStrategy

Pushback by adding nan_penalty * counter to the loss if NaNs are encountered.

The counter indicates how many NaNs occurred in a row. The nan_tolerance is the upper limit, if this is exceeded, the fallback will be used and an error is raised.

Parameters

• nan_penalty (Union[float, int]) – Value to add to the previous loss in order to penalize the step taken.

• nan_tolerance (int) – If the number of NaNs encountered in a row exceeds this number, the fallback is used.

minimize_nan(loss, params, minimizer, values=None)

Return type

float

zfit.minimizers.baseminimizer.DefaultStrategy

alias of zfit.minimizers.baseminimizer.PushbackStrategy

class zfit.minimizers.baseminimizer.DefaultToyStrategy(nan_penalty=100, nan_tolerance=30, **kwargs)

Bases: zfit.minimizers.baseminimizer.PushbackStrategy, zfit.minimizers.baseminimizer.ToyStrategyFail

Same as DefaultStrategy, but does not raise an error on full failure, instead return an invalid FitResult.

This can be useful for toy studies, where multiple fits are done and a failure should simply be counted as a failure instead of rising an error.

Pushback by adding nan_penalty * counter to the loss if NaNs are encountered.

The counter indicates how many NaNs occurred in a row. The nan_tolerance is the upper limit, if this is exceeded, the fallback will be used and an error is raised.

Parameters

• nan_penalty (Union[float, int]) – Value to add to the previous loss in order to penalize the step taken.

• nan_tolerance (int) – If the number of NaNs encountered in a row exceeds this number, the fallback is used.

minimize_nan(loss, params, minimizer, values=None)

Return type

float

class zfit.minimizers.baseminimizer.BaseMinimizer(name, tolerance, verbosity, minimizer_options, strategy=None, **kwargs)

Bases: zfit.minimizers.interface.ZfitMinimizer

Minimizer for loss functions.

Additional minimizer_options (given as **kwargs) can be accessed and changed via the attribute (dict) minimizer.minimizer_options.

property tolerance

step(loss, params=None)

Perform a single step in the minimization (if implemented).

Parameters

params (Union[Iterable[ZfitParameter], None, Iterable[str]]) –

Returns:

Raises

MinimizeStepNotImplementedError – if the step method is not implemented in the minimizer.

minimize(loss, params=None)

Fully minimize the loss with respect to params.

Parameters

• loss (ZfitLoss) – Loss to be minimized.

• params (Optional[Iterable[ZfitParameter]]) – The parameters with respect to which to minimize the loss. If None, the parameters will be taken from the loss.

Return type

FitResult

Returns

The fit result.

copy()

zfit.minimizers.baseminimizer.print_params(params, values, loss=None)

zfit.minimizers.baseminimizer.print_gradients(params, values, gradients, loss=None)