Using apprentice to perform function optimization

In apprentice, we can perform optimization on predition functions so as to minimize the loss function For this purpose, a base class Minimizer is provided in apprentice.

In this tutorial, we describe the Minimizer base class, the available implementation of the Minimizer base class, and how you can create your own implementation of the Minimizer base class. More specifically, in this tutorial, we will:

Test the install
Learn about the Minimizer base class
Learn about the available implementation of the Minimizer base class
Learn how to construct a different implementations of the Minimizer base class

Getting started

To install apprentice, execute the following commands:

git clone git@github.com:HEPonHPC/apprentice.git
cd  apprentice/
pip install .
cd ..

Then, test the installation as described in the test installation documentation.

`Minimizer` base class

Here, the functions of the Minimizer base class are discussed. More details can be found in minimizer code documentation.

Construction methods

To construct a Minimizer object, call the __init__ function:

def __init__(self, function, **kwargs)

where function is the object of a class that inherits the Function base class.

Abstract (unimplemented) methods

Minimizer has one abstract (unimplemented) method.

def minimize(self,x0): Minimize the prediction function with starting point x0. This function is required to be implemented in the inheriting class.

Available implementations of the `Function` base class in apprentice

ScipyMinimizer

The implementation of the minimize function in ScipyMinimizer minimizes the prediction function using SciPy. The signature of the abstract function is:

def minimize(self, x0=<array>, nrestart=<int>, method=<str>, tol=<float>)

where

x0: starting point
nrestart: starting point
method: solver (method) to use. The allowed values include
- “tnc”: Truncated Newton (TNC) algorithm
- “lbfgsb”: L-BFGS-B algorithm
- “slsqp”: Sequential Least Squares Programming (SLSQP)
- “ncg”: Newton conjugate gradient trust-region algorithm
tol: tolerance value

See ScipyMinimizer code documentation for more details. Additionally, the ScipyMinimizer unit test script contains the construction and usage of the operations over the ScipyMinimizer object.

Construct your own implementation of the `Minimizer` base class

To implement your own minimizer (optimization) function, all you have to do is to implement the abstract function. Then you can construct your object using the construction methods. To override the __init__ constructor method, use the template in the code snippet below:

def __init__(self, function, **kwargs):
    super(<Your class name>, self).__init__(function, **kwargs)
    """
    add additional construction code here
    """
    # ...