SNoW's -interactive mode has a slightly different example file format. Associated with each example in this mode is a single character which instructs SNoW to take a particular action on this example. Currently, there are three possible actions:
Syntax:
For example, the following line requests SNoW to evaluate the example with active features 1001, 1005, 1009, 1010, and 1200.
e, 1001, 1005, 1009, 1010, 1200:
The output format depends on the value of parameter -o. If the value is accuracy, which is the default, then no output will be given.
Syntax:
For example, the following line asks SNoW to promote targets 1, 3, and 5 for the example with active features 1002, 1013, 1018, and 1011.
p, 1002, 1013, 1018, 1011; 1, 3, 5:
Similarly, the following line demotes targets 2 and 4 for the example with active features 1004, 1007, 1212, 1311, and 1312.
d, 1004, 1007, 1212, 1311, 1312; 2, 4:
These two training control commands won't generate any output.