How does tMatchPredict predict values on a dataset?
Once the learning model is built, the tMatchPredict component can
predict values on a dataset using the model it receives from the
tMatchModel component.
The input records can be either paired or unpaired:
- If the input records are paired, the tMatchPredict component can label suspect duplicates automatically.
- If the input records have not been paired, use the pairing model generated by the tMatchPairing component to compute the pairs of suspect duplicates.
Rather than returning pairs, the component can also return groups of records that are matching one another, by adding a clustering step in the algorithm. You can define the clustering classes, which are generally the label corresponding to a match.
The algorithm used for clustering computes connected components of the graph where each vertex is a record. There is an edge between two vertices if the pair of records has the right label.
For example, if record A matches record B and record B matches record C, a group including records A, B and C is created even if record A and record C do not match.