DESCRIPTION
This program is used to manage user-supplied hypotheses and evidences in a Dempster-Shafer knowledge
base file. The knowledge base file is an XML format ASCII file which resides in its own 'DST' element
inside the user's current mapset.
Such a file is necessary to perform Dempster-Shafer Theory (DST) calculations using
dst.combine .
Although the knowledge base file can theoretically be edited with
any ASCII editor, it is highly recommended to always use dst.update to make any changes
to it. This ensures that the mathematical structure of the model layout will always remain intact.
If you do not know what DST is or why you would need this program, please refer to the
dst.combine command's manpage for an introduction.
The dst.update command allows the user to register hypotheses in a knowledge base file
and attach evidences which support these hypotheses.
To get an overview of the DST model structure at any time,
use the dst.view command.
Knowledge base files themselves can be listed, created, deleted and renamed using a specialised
set of commands.
Registering Hypotheses
As described in the dst.combine command's manpage,
the user-suppplied hypotheses (type 'USER') form the backbone of any DST analysis. The Frame of Reference (FOD)
is the set of hypotheses that cover all possible outcomes of the analysis. In the DST, this also
includes two types of hypotheses, which the user does not have to supply.
- The NULL hypotheses, which always exists.
- Sets of all possible combinations of user-supplied singleton hypotheses, which exist if more than one
hypothesis is supplied.
It is vital that the FOD is always complete and that the hypotheses are stored in the correct
hierachical order. E.g., if three hypotheses {a},{b} and {c} were supplied, one after another,
the correct and complete FOD would look like:
{NULL}
{a}
{b}
{a,b}
{c}
{a,c}
{b,c}
{a,b,c}
The order in which {a},{b} and {c} are given has no influence on the correctness of the DST
calculations. Nevertheless, keeping the correct structure intact can be challenging if the
FOD is subject to changes. With dst.update, the user can add, rename and remove
hypotheses in the FOD, without having to worry about its structure.
In fact, the user only adds, renames are removes singleton hypotheses (except {NULL})
, sets are always managed automatically and stored as type 'AUTO' in the knowledge base file.
According to the laws of combinatorics, the number of hypotheses sets in the FOD grows rapidly
if more hypotheses are supplied by the user. Currently, up two 12 user-supplied hypotheses
are supported, but it seems unlikely that DST calculations with more than 4-6 user hypotheses
will finish within reasonable time, as processing time also grows exponentially.
You can manage user-supplied hypotheses by using the add=[hyp], del=[hyp],
ren=[hyp] (plus new=[hyp]) parameters.
All other hypotheses will automatically be updated in the correct way. Note that names of
hypotheses, like filenames, must not contain spaces or special characters. They will also
be treated case-sensitive so 'HYP' is a different hypothesis than 'Hyp'.
Attaching Evidences
Once all hypotheses have been specified (it is vital for correct results that no possible
hypothesis be left out!), it is up to the user to specify which evidences support a given
hypothesis (or set of hypotheses).
Any evidence must be the result of a correct BPA calculation (see
r.dst.bpa for more details on this).
Different GRASS map formats can be attached using the const=[name], rast=[name},
vect=[name] and site=[name] parameters.
However, currently only GRASS raster maps or constant values can be used to
supply evidence. Handling of other vector evidence is not yet implemented in the DST combination
routines themselves.
Thus, only evidence attached using the rast=[name] parameter will produce output
if you call the
dst.combine command. Evidence of type 'const'
consists of a single floating point number. It can be useful for testing purposes and to
get a feeling for what the DST functions do with your data.
If you need to remove erraneous or outdated evidence from a knowledge base file, you can use
the clean=[hypothesis] and prune=[evidence] parameters (see description
of parameters below).
If all hypotheses and evidences are present in the data base, they have to be grouped
to define sources of evidence. This can be done using the
dst.source command.
Parameters
- file=name
- Name of DST knowledge base file to modify. You can get a listing of DST
knowledge base files in your current mapset by using
dst.list .
- add=name
- Adds a hypothesis 'name' to the knowledge base. Naming conventions follow
the same rules as filenames.
- del=name
- Deletes hypothesis 'name' (and all references in hypotheses sets) from the
knowledge base file.
- ren=name
- Renames hypothesis 'name' (and all references in hypotheses sets) from the
knowledge base file. New name must be specified using the new=[name]
parameter (see below).
- new=name
- Specifies a new name for renaming a hypothesis.
- const=value
- Attaches a constant value as evidence to a hypothesis. As with all evidences,
the target hypothesis must be specified using the hypothesis=[name]
parameter.
- rast=name
- Attaches a GRASS raster map as evidence to a hypothesis.
- vect=name
- Attaches a GRASS vector map as evidence to a hypothesis (useless for now).
- clean=name
- Removes all attached evidence from a hypothesis (specified using hypothesis=[name]).
- prune=name
- Removes all references to an evidence from one or all hypotheses. The type(s) of
evidence to detach can be specified using type=[name].
- hypothesis=name
- Specifies hypothesis to operate on. Note that order matters when refering to
hypotheses sets: 'a,c' may exist, but 'c,a' not!
- type=name
- Type of evidence to detach (use in conjunction with prune=[hyp]).
Possible options are const,rast,vect,*.
- precision=value
- Number of decimal places for storing evidence of type const. The default is '2'.
USAGE
Refer to the section on 'Predictive Modelling' in the
manpage of the dst.combine command
for an example of how to register hypotheses and evidences using dst.update.
Notes
This program was developed as part of the GRASS 5 DST Predictive Modelling Toolkit.
The term 'knowledge base file' was first used in the manual of the
IDRISI GIS software which also offers some DST functionality.
SEE ALSO
dst.combine
dst.list, dst.create, dst.remove, dst.rename
dst.source
dst.view
r.dst.bpa
AUTHOR
Benjamin Ducke,
University of Kiel, Germany
Last changed: 2004/07/21