ObjectOriented Minimization

1
Object-Oriented Minimization

• David Sachs, Mark Fischler
• What is the OO Minimization project?
• Why is it worth doing?
• Some alternative related work
• Concepts and techniques
• The recent package review
• Status and intentions

2
What is the OO Minimization project?

• C so as not to tie to FTN
• User functions are mostly in C nowadays
• Under the ZOOM umbrella
• All functionality of Minuit
• plus completions orthogonal natural concepts
• e.g. domains limited on only one side
• Same ground rules and assumptions as Minuit
• Highly efficient when cost of function call is
  high
• Not an f2c of MINUIT
• that would forfeit all the O-O advantages
• Factorization of concepts
• clarifies what is going on when reading code

3
Who is OO Minimization for?

• Stakeholders are the whole HEP community
• Much of the minimization in our community is used
  in the context of data fitting
• Replacements for the venerable Minuit are
  becoming necessary
• Three reactions from physicists
• Are you out of your mind nobody needs a
  replacement for Minuit
• Are you out of your mind this must already
  have been done
• But in fact, a truly object oriented Minimization
  package has not been done, and is not happening
  except for this one
• Are you out of your mind why didnt you
  provide this years ago?

4
Object-Oriented Minimization

• What is the OO Minimization project?
• Why is it worth doing?
• Some alternative related work
• Concepts and techniques
• The recent package review
• Status and intentions

5
What are the advantages of this package?

• Re-grounding and cleanup
• Clear expositions of the algorithms, definitions
  and assumptions of Minuit
• Supplying some related capabilities that fill
  obvious gaps
• Correction of several (minor) errors in existing
  Minuit
• Extensibilty
• Easy to add new termination criteria, domain
  types
• Straightforward to add new algorithms
• Goal is that the work to add a new algorithm is
  dominated by the work of developing the
  algorithm coordinating it within the minization
  framework should be routine.

6
Why is O-O Minimization Desirable?

• The O-O style has advantages which are relevant
  for the way we want to do minimization
• The function to be minimized can be related to
  an object with state information
• A problem can be treated as a true object
• Multiple simultaneous minimizations in a job

7
Object-Oriented Minimization

• What is the OO Minimization project?
• Why is it worth doing?
• Some alternative related work
• Concepts and techniques
• The recent package review
• Status and intentions

8
Alternative Related Work

• People have done and are doing ports of Minuit to
  something that will compile under C compilers
• ROOT does this for its fitter
• Fred James and Winkler are creating a package
  that will adhere closely to Minuit
• Not really object oriented
• Have a much less ambitious philosophy hard to
  actively collaborate without sacrificing our
  goals
• There are various minimizers for fitters
• Fitter applications can be coded without
  considering a minimization problem as an
  independent object, so these packages all forego
  the advantages of O-O.
• We are betting that if a high-quality minimizer
  with good O-O design becomes available, it will
  fit the style of programming that the HEP
  community will be moving toward, and see
  extensive use.

9
Object-Oriented Minimization

• What is the OO Minimization project?
• Why is it worth doing?
• Some alternative related work
• Concepts and techniques
• The recent package review
• Status and intentions

10
Concepts and Techniques

• The Function
• Can by a global function
• Can also be a functor
• The Domain
• Expresses restrictions on the solution space
• Expresses maps to implement these restrictions
• The Problem
• Coordinates application of steps of an algorithm
  to minimize a function in a domain

11
Dynamic Concepts

• The Algorithm
• Generic for methods that move to a superior point
  in terms of minimizing the function
• The package includes all the Minuit algorithms
  (MIGRAD, SIMPLEX, ) and will expand to include
  others.
• Termination Criteria
• Function call limits
• Estimated distance to minimum criteria
• Uncertainty in parameter values criteria
• Termination, Domain, Algorithm and Function are
  all orthogonal concepts
• expressed by distinct classes

12
Object-Oriented Minimization

• What is the OO Minimization project?
• Why is it worth doing?
• Some alternative related work
• Concepts and techniques
• The recent package review
• Status and intentions

13
The Recent Package Review

• JBK and MP have just completed a Minimization
  Package Review
• Is this package easy to use
• Does the class decomposition make sense
• Is the library extensible
• Is the library maintainable

14
The Package Review Comments

• There were a fair number of shortcomings related
  to producing a truly object-oriented minimization
  scheme
• many types of objects would have behavior which
  users will find annoying if they really want to
  exploit the advantages of O-O code
• despite intentions in the other direction, we
  have been overly influenced by the way minuit
  does things and views concepts
• The package was describe as currently being
  object-centric rather than object oriented

15
Package Review Suggestions

• The review cited a need for interfaces to support
  interactive minimization an interactive tool
• The package should not tie in to a particular GUI
  or command line interface, but should provide
  those methods which will commonly be needed by
  those
• There were the usual collection of suggestions
  for
• better encapsulation
• decoupling improving cohesion
• cleaning up interfaces

16
Package Review Reaction

• We are going through the review document and
  formulating action items to implement their
  suggestions
• Currently about 25 items, mostly small
• Will end up with close to 40
• We will prioritize the items (most are worth
  doing) and start on implementation
• Actually, while the perusal of the document goes
  on, David has been taking care of some of the
  action items

17
Object-Oriented Minimization

• What is the OO Minimization project?
• Why is it worth doing?
• Some alternative related work
• Concepts and techniques
• The recent package review
• Status and intentions

18
Status and Intentions

• The package is in usable shape
• We will be getting in touch with friendly users
  to let them try it out
• with the understanding that the interface will
  change a bit in reaction to the review
• We will be releasing the package once most of
  the review suggestions are dealt with
• By that time, ZOOM packages will standardly be in
  a form suitable for general HEP (non-Run II) use

19
Intended Extensions

• Sample interactive minimization tool
• We want to be able to verify that the interface
  for interactive tools is suitable
• Use cases of interacting with outside packages
• R statistical package using minimizer
• ROOT-involved example
• Termination Criteria Collection
• Superior documentation
• Gather into good technical documentation the work
  done in studying the algorithms
• Significantly more mantra-illustrating examples
• Try out mantras for extension
• Add a new Algorithm
• Create a new type of Domain

20
Time-scales

• In a couple of months DS e MF
• can hope to be at point of releasing package with
  the improvements
• By CHEP time (late September)
• Should be seriously talking with James about how
  to avoid dichotomy of packages in the HEP
  community
• But should not cave in on our goals for the sake
  of adhering more closely to old style
  (procedural) coding philosophy

21
(No Transcript)