The purpose of this appendix is to describe the Decision Table Visualizer's data and configuration files. The *.dtableviz file contains a schema and optional history section, and the *.data file contains the data. The format of these two files is almost exactly that described in Appendix A.
The Decision Table Visualizer's *.dtableviz file must contain a schema of the form: some number of columns of type string or enum; followed by a float column containing the weight of records; followed by a vector column, indexed by an enum containing possible classes, and containing the proportion of the weight of records in each class. The *.dtableviz and *.data files are automatically generated by the Tool Manager, and you should not normally need to modify them.
There are two extra key-words that the schema in the *.dtableviz file uses that are not present in other schemas. They are "auto", and "nominal". The Decision Table Visualizer must be able to differentiate between different types of enums. For example, enums that have been automatically rendered discrete by mining are handled differently in drill-through, than enums that have been binned by the user in Tool Manager. Hence the keyword "auto" is used to distinguish columns that have been binned by mining.
The label may be a binned column, or it may be string column. Either way it appears as an enum in the schema, so that the final probs[] column can be indexed by it. If the label is string valued, the "enum" keyword is predicated with "nominal" to distinguish it.
An example configuration file created from the adult dataset follows:
MineSet 2.5
input {
file "adult-tmbin-dtab.dtableviz.data";
enum string `hours_per_week_bin_k`{"- 20", "20-28", "28-36",
"36-44", "44-52", "52-60", "60+"};
enum `hours_per_week_bin_k``hours_per_week_bin`;
auto enum string `gross_income_k`{"- 9598", "9598-14579",
"14579-24794.5", "24794.5-24806", "24806-29606", "29606-42049.5",
"42049.5-46306.5", "46306.5-64885", "64885+"};
enum `gross_income_k``gross_income`;
auto enum string `final_weight_k`{"- 223033", "223033+"};
enum `final_weight_k``final_weight`;
float `weight`;
nominal enum string `label`{"Female ", "Male "}; # the
label values
float `probs[]`[ enum `label`]; # and probabilities
}
history {
:
:
}
|
In this example, the name of the data file is adult-tmbin-dtab.dtableviz.data. The first three columns are attributes from the data. The first one, hours_per_week_bin has been binned by the user using Tool Manager. The second two, gross_income, and final_weight were binned automatically by the mining. Note the use of the auto keyword in their enum definition. The fourth column gives the weight of records having the values given by the first three columns.
The last column, probs[], is a vector valued column which gives the probability of each class, computed by dividing the number of records for each class by the value in the weight column. The sum of the entries of this vector must add to 1. Note that the definition of the index label includes the keyword nominal to show that the class values are not numeric, and hence do not have an implied ordering. The nominal keyword would not be present had the label been a binned numeric attribute.