The class GeoTableEval provides a tool for evaluating multiple partitions
of fundamental geographical area definitions called regions into aggregations. All partitions are
evaluated in terms of a selected error metric.
The input
data for the tool should consist of (a) boundary points of regions and (b) attributes
associated with each region.
This type of data can be loaded from ESRI Shape files (.shp, .shx and .dbf files).
Description:
The objective of this tool is to evaluate feature prediction errors per each aggregation of regions
based on one or many partitions of regions. Feature prediction errors are computed based on a selected set
of partitions, features and associated feature weights.
Setup:
It is assumed that a shape file (.shp and .shx files) with all boundary and tabular information
has been loaded in the main window. When the GeoTableEval dialog is invoked
from the main menu of the main window, the file with tabular information (.dbf) called the attribute file
will be loaded assuming that the root name of the shape file and attribute file are identical.
All loaded data are represented by two internal data objects,
ShapeObject (region boundaries) and Table (region attributes/features, weights, aggregation labels, auxiliary information).
The region boundaries are visualized in the main window and the region attributes
can be visualized from the GeoTableEval dialog by clicking on the "ListFeat" or "ShowDbf" button (see Dialog Figure).
An example of region boundaries representing the county boundaries in Illinois.
An example of region attributes in a tabular form that represent attributes per each
region boundaries, e.g., terrain elevation attributes of counties in Illinois.
The Dialog interface to perform parameter setup, evaluation, visualization
and output of the computed results.
Step 1:
The first step is to select aggregations (partitions), attributes (features) and
feature weights before performing any error evaluations. One can view labels for all
tabular columns by clicking the button "ListFeat" and select any combination
of aggregations, features and feature weights by clicking on "SelectAggreg",
"SelectFeat" and "SelectWeights" and entering column indices according to the
list of column labels. For example, if a user would want to select
attributes/features 1, 2, 3, 4, 6, 7, 8 and 10, then he/she would type
into a pop-up dialog 1-4, 6-8, 10, or each index separated by a space or comma.
After selecting a set of aggregations, features and weights,
the TextArea labeled as "Results" in the dialog will show the list of selected column
labels.
Note 1: There will be a warning issued if a column index selected by "SelectAggreg"
is not of integer type since we expect to process aggregations labeled by integer
values.
Note 2: If the feature weights are not selected then the error evaluation is performed with weights
set to 1.0. If the feature weights are selected as a single number then this value
will be applied to all features. Any other selection of feature weights must insure
the same number of weights and features otherwise the evaluation will fail.
Step 2: Select an error metric from a combo box "EvalMetric". The metrics are described next.
Run and Display Results: The algorithm is executed by clicking the "Eval" button. The global errors of all partitions can be viewed by clicking "ShowEval" and viewing the TextArea labeled as "Results".
Furthermore, a plot showing computed errors as a function of region (boundary index) will be displayed (see below). Error values for multiple partitions are distinguished with different colors. It is possible to change the horizontal axis labeling by choosing a column in the attribute file that describes each region. For example, by clicking the "SelectBoundID" button, a user can choose county indices from an attribute file for the horizontal axis rather than a row index of the region in the attribute file.
The error evaluations can also be displayed by clicking the button "ShowGeoEval". This type of visualization presents the error of each aggregation together with the information about geographical locations of individual regions within an aggregation. By selecting a visible band (see the main documentation about viewing images), one can view geographical error distribution for each evaluated partition. Results for each partition can be viewed separately by invoking the "BandSelect" Dialog after right mouse button click on the image area and selecting the index of the partition. It is also possible to view the results as a movie by sweeping through the error layers continuously. This type of visualization is enabled with the "PlayBands" Dialog that is also invoked by right mouse button click on the image area.
Output of Results:
The error evaluations can be saved into a DBF file by clicking the "Insert2DBF" button and
then the "SaveDBF" button. It is also possible to save the results shown by "ShowGeoResults" with the button
"SaveGeoResults".
It might be desirable to view the loaded attributes with the inserted labels by clicking "ShowDBF".
This will open a new table viewer with all dbf file column attributes and their values.