Like the above, but just for one source node at a time.
This one is similar to selectNegativeExamples, but instead of looking specifically for _training_ examples that are close to the given positive examples, we look across the whole domain and range of a relation to find things to score.
This one is similar to selectNegativeExamples, but instead of looking specifically for _training_ examples that are close to the given positive examples, we look across the whole domain and range of a relation to find things to score. The point of this is to actually perform KB completion, instead of just training a model or doing cross validation. So this method is used to generate possible predictions for NELL's ongoing run, for instance.
Returns a new Dataset that includes the input data and negative instances sampled according to PPR from the positive examples in the input data.