usecase: Classify data

public usecase: Classify data
Author: Lukáš Civín
Project: Phase: 1.0; Status: Proposed; Version: 1.0; Complexity: 1
Dates: Created: 5.12.2005 16:56:11; Modified: 6.12.2005 10:58:07;
Flags: Active: false; IsRoot: false; IsLeaf: false;
Extension Points:
UUID: {F37B6528-DCEA-4dbc-882B-7F376B090E8A}
Menu: Action>Classify
 

Goto: Constraints, Scenarios

See also: User, Select Input Columns for BP network, Select output columns for BP network, Dialog for rounding output, Insert name of a new version

Appears in: UC Classification

Connections
 
Classify data Constraint
Constraint Type Status Detail
The net is selected Pre-condition Approved
 
Classify data Scenarios
Scenario Type Detail
Classification with BP network Basic Path When the net is trained, it is used to classify new data. So for each input data, the network "answer" classification - output data.

It uses some processes from training network
- GetRow()
- FillDoubles() - inputs and outputs

goal function Result() returns the double [] with outputs of the network. It uses function TraceForward(), well known from Training network.

After all a new row is created - a row of the table - with output values, input values (that has not been overwritten by output values) and unused values. If there is output and input column the same, output value will overwrite it.

Example: The user has 5 columns - input columns chooses 1,2,3. Output column - selects 3,5 and newly created 6 (that's why a new structure for classification output columns is used - after cancelling classification, this output structure, with potential new columns, must be forgotten). When data flows through the network - these values are written:
-output values into columns 3,5,6
-input values into columns 1,2
-unused value into column 4
New table is created with different structure (6th column)