Table 12 Results: Codes with Less Than Five Positive Documents. We see that performance improves as the number of positives improves. The row labeled GT4 shows the performance for these codes when they are processed in the standard way using all available positive documents. The test set represented in the FirstFiveTest column is limited to documents time stamped on or earlier to the time stamp of the fifth positive document that is not in the training set. The test set in the FullTest column includes the rest of the temporal stream following the training set.

From: GO for gene documents