Re: How much requirement documentation up front?

From: H. S. Lahman (h.lahman_at_verizon.net)
Date: 10/27/04 

Date: Wed, 27 Oct 2004 16:04:56 GMT

Responding to Parker...

>>>>As far as the evolution of standards is concerned, that is a quite
>>>>different situation because the subject matter itself is being defined.
>>>>
>>>
>>>But that's sometimes a characteristic of business applications as well.
>>
>>Do you have an example? I am really curious
>>because I spent a decade doing core IT systems and I can't think of one.
>>
>
> Most of the financial trading system and risk management projects I've been
> involved with have had a significant business learning component that took
> place over the life of the project. Much of this learning component was
> related to the application of numerical methods to complex valuation and
> simulation problems, where the expertise was distributed between
> quantitative analysts and developers like myself who made this area
> something of a speciality. In general, any project that has some R&D
> component will have this characteristic, to more or lessor extent.

OK, see what you meant but I think we draw the line in the sand for
requirements analysis differently. B-)

When I was doing inventory control we had a similar problem with
forecasting algorithms. [It was for a food retailer where if one
doesn't have a complete inventory turn in less than 10 days, one goes
broke. So it is critical to get the forecasts correct at the store
level -- essentially JIT at the store with a warehouse buffer to absorb
store variance.] A priori one does not know which algorithm will
provide a best fit for a particular product. So one has to do some
investigation with fits to historical data. But I would consider that
part of requirements analysis; one is just using software for the fit
analysis prior to implementing the IC system.

The interesting aspect from your perspective is related to new products
where there is no historical data or products where consumer habits
change. The former is usually done by matching to category data and
then monitoring until one has enough historical data to determine
variance. The latter is done by periodically evaluating the variance to
forecast. When one gets to an unacceptable variance, one does the
fitting analysis again and selects a new algorithm.

In either case, the monitoring is just using the IC system as a data
collection tool for requirements analysis. The need to change the
forecasting algorithm was driven by detecting an unacceptable variance.
  So one goes through the fit analysis again to identify the new best
fit algorithm that fit the new requirements.

Though this is an iterative, ongoing process that involves the software
product's results, I don't really see it as an argument for guessing at
requirements in the software product itself. You would regard the best
category fit for a new product as a guess at the requirements. I don't
see that as a guess; one is specifying the algorithm requirement as
completely, precisely, and unambiguously as one can at the time and
there is no guessing about alternatives.

The fact that the software product itself makes it very convenient to
log the necessary historical data for testing variance and requirements
analysis doesn't mean that one is guessing at the requirements for the
product itself. The analysis of the historical data that determines
which algorithm to use is still a separate activity from the business of
the IC system that I would regard as requirements analysis. So any
change to the product algorithm used by the IC systems is the result of
repeating that entire activity with different results (i.e., the
requirements _for the software product_ have changed).

[In practice the fitting analysis software is likely to be a module
within the IC system. But that module is an analysis tool that is
independent of the business of actually managing the inventory on a
day-to-day basis. That is, it is a separate R&D tool from the IC system
whose requirements are being specified. I have no doubt that today's IC
systems (I did mine in the '70s) are self-modifying in that they
automate the analysis process and modify the forecasting algorithms over
time without developer intervention. But that sort of system is a whole
different set of requirements. B-)]

*************
There is nothing wrong with me that could
not be cured by a capful of Drano.

H. S. Lahman
hsl@pathfindermda.com
Pathfinder Solutions -- Put MDA to Work
http://www.pathfindermda.com
blog (under constr): http://pathfinderpeople.blogs.com/hslahman
(888)-OOA-PATH

Relevant Pages

  • Re: C++: why 80 charachters??
    ... I guess if the result of your editor's word-wrapping algorithm is ... to be able to fit more than about 130 characters on my ... print utility to do ...
    (comp.lang.cpp)
  • Re: random model
    ... My GA algorithm is in fact searching for coefficients to fit a dataset. ... > thinking of are inherent in the GA itself, whatever you're modeling; ...
    (sci.math)
  • Re: 3-D fit for extrapolation
    ... of equations in as many unknowns as you have nans. ... As for evaluating a polynomial fit over the entire ... with the simplex algorithm, especially since there ...
    (comp.soft-sys.matlab)
  • Re: algorithm to sum run time of simultaenous jobs?
    ... jobs are a match for the size of the objects you're trying to fit into ... the 5 bins? ... then my understanding is that there is no algorithm outside of ...
    (rec.puzzles)