Parameter#

class gammapy.modeling.Parameter(name, value, unit='', scale=1, min=nan, max=nan, frozen=False, error=0, scan_min=None, scan_max=None, scan_n_values=11, scan_n_sigma=2, scan_values=None, scale_method='scale10', interp='lin', prior=None, scale_transform='lin')[source]#

Bases: object

A model parameter.

Note that the parameter value has been split into a factor and scale like this:

value = factor x scale

Users should interact with the value, quantity or min and max properties and consider the fact that there is a factor and scale an implementation detail.

That was introduced for numerical stability in parameter and error estimation methods, only in the Gammapy optimiser interface do we interact with the factor, factor_min and factor_max properties, i.e. the optimiser “sees” the well-scaled problem.

Parameters:

namestr

Name.

valuefloat or Quantity

Value.

scalefloat, optional

Scale (sometimes used in fitting).

unitUnit or str, optional

Unit.

minfloat, str or quantity, optional

Minimum (sometimes used in fitting). Default is None, which set the attribute to np.nan.

maxfloat, str or quantity, optional

Maximum (sometimes used in fitting). Default is None, which set the attribute to np.nan.

frozenbool, optional

Frozen (used in fitting).

errorfloat

Parameter error.

scan_minfloat

Minimum value for the parameter scan. Overwrites scan_n_sigma.

scan_maxfloat

Maximum value for the parameter scan. Overwrites scan_n_sigma.

scan_n_values: int

Number of values to be used for the parameter scan.

scan_n_sigmaint

Number of sigmas to scan.

scan_values: `numpy.array`

Scan values. Overwrites all the scan keywords before.

scale_method{‘scale10’, ‘factor1’, None}

Method used to set factor and scale. Default is “scale10”.

interp{“lin”, “sqrt”, “log”}

Parameter scaling to use for the scan. Default is “lin”.

priorPrior

Prior set on the parameter.

scale_transform{“lin”, “sqrt”, “log”}

Parameter scaling. Default is “lin”.

Attributes Summary

conf_max	Confidence maximum value as a float.
conf_min	Confidence minimum value as a float.
error
factor	Factor as a float.
factor_max	Factor maximum as a float (used by the optimizer).
factor_min	Factor minimum as a float (used by the optimizer).
frozen	Frozen (used in fitting) (bool).
max	Maximum as a float.
min	Minimum as a float.
name	Name as a string.
prior	Prior applied to the parameter as a Prior.
quantity	Value times unit as a Quantity.
scale	Scale as a float.
scale_method	Method used to set factor and scale.
scale_transform	scale interp : {"lin", "sqrt", "log"}
scan_max	Stat scan maximum.
scan_min	Stat scan minimum.
scan_n_sigma	Stat scan n sigma.
scan_values	Stat scan values as a ndarray.
type
unit	Unit as a Unit object.
value	Value = factor x scale (float).

Methods Summary

autoscale()	Apply `interpolation_scale' and `scale_method' to the parameter.
check_limits()	Emit a warning or error if value is outside the minimum/maximum range.
copy()	Deep copy.
inverse_transform(factor)	Inverse tranform from factor (used by the optimizer) to value.
prior_stat_sum()
reset_autoscale()	Reset scaling such as factor=value, scale=1.
set_lim([min, max])	Set the min and/or max value for the parameter.
to_dict()	Convert to dictionary.
transform(value[, update_scale])	Tranform from value to factor (used by the optimizer).
update_from_dict(data)	Update parameters from a dictionary.
update_scale(value)	Update the parameter scale.

Attributes Documentation

conf_max#

Confidence maximum value as a float. Return parameter maximum if defined, otherwise a default is estimated from value and error.

conf_min#

Confidence minimum value as a float. Return parameter minimum if defined, otherwise a default is estimated from value and error.

error#

factor#

Factor as a float.

factor_max#

Factor maximum as a float (used by the optimizer).

By default when no transform is applied, factor_max = max / scale, otherwise factor_max = transform(max).

factor_min#

Factor minimum as a float (used by the optimizer).

By default when no transform is applied, factor_min = min / scale, otherwise factor_min = transform(min).

frozen#

Frozen (used in fitting) (bool).

max#

Maximum as a float.

min#

Minimum as a float.

name#

Name as a string.

prior#

Prior applied to the parameter as a Prior.

quantity#

Value times unit as a Quantity.

scale#

Scale as a float.

scale_method#

Method used to set factor and scale.

scale_transform#

scale interp : {“lin”, “sqrt”, “log”}

scan_max#

Stat scan maximum.

scan_min#

Stat scan minimum.

scan_n_sigma#

Stat scan n sigma.

scan_values#

Stat scan values as a ndarray.

type#

unit#

Unit as a Unit object.

value#

Value = factor x scale (float).

Methods Documentation

autoscale()[source]#

Apply `interpolation_scale’ and `scale_method’ to the parameter.

check_limits()[source]#

Emit a warning or error if value is outside the minimum/maximum range.

copy()[source]#

Deep copy.

inverse_transform(factor)[source]#

Inverse tranform from factor (used by the optimizer) to value.

Parameters:

valuefloat

Parameter factor

prior_stat_sum()[source]#

reset_autoscale()[source]#

Reset scaling such as factor=value, scale=1.

set_lim(min=None, max=None)[source]#

Set the min and/or max value for the parameter.

Parameters:

min, max: float, `~astropy.units.Quantity`, str

Minimum and Maximum value to assign to the parameter min and max. Default is None, which set min and max to np.nan.

to_dict()[source]#

Convert to dictionary.

transform(value, update_scale=False)[source]#

Tranform from value to factor (used by the optimizer).

Parameters:

valuefloat

Parameter value

update_scalebool, optional

Update the scaling (used by the autoscale). Default is False.

update_from_dict(data)[source]#

Update parameters from a dictionary.

update_scale(value)[source]#

Update the parameter scale.

Set factor and scale according to scale_method attribute.

Available scale_method.

• scale10 sets scale to power of 10, so that abs(factor) is in the range 1 to 10

• factor1 sets factor, scale = 1, value

In both cases the sign of value is stored in factor, i.e. the scale is always positive. If scale_method is None the scaling is ignored.