Parameter¶

Several objects in zfit, most importantly models, have one or more parameter which typically parametrise a function or distribution. There are two different kinds of parameters in zfit:

Independent: can be changed in a fit (or explicitly be set to fixed).
Dependent: cannot be directly changed but _may_ depend on independent parameters.

Unique names¶

Parameters in zfit are global, unique objects. No Parameters with the same name can therefore exist as its meaning would be ambiguous. If a new parameter with the same name will be created, a NameAlreadyTakenError will be raised.

For Jupyter notebooks, see also :ref:`Parameters in Jupyter <parameters_in_jupyter>`

Independent Parameter¶

To create a parameter that can be changed, e.g., to fit a model, a :py:class~zfit.Parameter` has to be instantiated.

The syntax is as follows:

param1 = zfit.Parameter("unique_param_name", start_value[, lower_limit, upper_limit])

Furthermore, a step_size can be specified. If not, it is set to a default value around 0.1. Parameter can have limits (tested with has_limits()), which will clip the value to the limits given by lower_limit() and upper_limit().

Note

Comparison to RooFit

While this closely follows the RooFit syntax, it is very important to note that the optional limits of the parameter behave differently: if not given, the parameter will be “unbounded”, not fixed (as in RooFit). Parameters are therefore floating by default, but their value can be fixed by setting the attribute floating to False or already specifying it in the init.

The value of the parameter can be changed with the set_value() method. Using this method as a context manager, the value can also temporarily changed. However, be aware that anything _dependent_ on the parameter will have a value with the parameter evaluated with the new value at run-time:

import zfit

mu = zfit.Parameter("mu_one", 1)  # no limits, but FLOATING (!)
with mu.set_value(3):
    print(f'before {mu}')

# here, mu is again 1
print(f'after {mu}')

before <zfit.Parameter 'mu_one' floating=True value=3>

after <zfit.Parameter 'mu_one' floating=True value=1>

Dependent Parameter¶

A parameter can be composed of several other parameters. They can be used equivalently to Parameter.

mu2 = zfit.Parameter("mu_two", 7)

def dependent_func(mu, mu2):
    return mu * 5 + mu2  # or any kind of computation
dep_param = zfit.ComposedParameter("dependent_param", dependent_func, params=[mu, mu2])

print(dep_param.get_params())

OrderedSet([<zfit.Parameter 'mu_one' floating=True value=1>, <zfit.Parameter 'mu_two' floating=True value=7>])

A special case of the above is ComplexParameter: it provides a few special methods (like real(), conj() etc.) to easier deal with complex numbers. Additionally, the from_cartesian() and from_polar() methods can be used to initialize polar parameters from floats, avoiding the need of creating complex tf.Tensor objects.

Parameters in Jupyter¶

Parameters are unique, global objects. This can conflict with the typical workflow in a jupyter notebook as cells are often executed multiple times. If a cell that creates a parameter is executed again (meaning a parameter with the same name as already existing should be created), it raises a :class:~`zfit.exception.NameAlreadyTakenError` (there is `an extensive discussion of the why <https://github.com/zfit/zfit/issues/186>`_)

To circumvent this, which comes from the fact that Jupyter is stateful, there are a few ways:

if possible, simply rerun everything.
move the creation of the variables into a separate cell at the beginning. Remember that you can set a value on a variable anytime using set_value which can be called as often as desired.
create a wrapper that returns the same parameter again if it exists. With this way it is clear what is done and it is convenient to use as a de-facto drop-in replacement for :class:~`zfit.Parameter` (using it in other places except for exploratory work may has unintended side-consequences)

Example wrapper:

all_params = {}

def get_param(name, *args, **kwargs):
    """Either create a parameter or return existing if a parameter with this name already exists.

    Args:
        name: Name of the Parameter
        *args:
        **kwargs:

    Returns:
        ``zfit.Parameter``
    """
    if name in all_params:
        return all_params[name]
    parameter = zfit.Parameter(name=name, *args, **kwargs)
    all_params[name] = parameter
    return parameter

Space, Observable and Range Building a model