Before you can begin initial training on the mastering model, you first create the unified dataset and configure the blocking model. Once a curator applies feedback and updates results at least once, then you can begin training the model.
To do this, on the Pairs page, start by labelling an initial set of pairs as a Match or No match . These initial responses provide the first feedback required to begin machine learning. Typically, one team member reviews these records and identifies obvious matching and non-matching pairs, as well as noting pairs that are difficult to assess.
In order for Tamr Core to learn best, provide a variety of data that represent the different ways in which records may or may not match. Use careful approaches to ensure Tamr Core has an accurate set of training data.
Remember, you can select Compare details to open a side-by-side view to compare records. See Viewing Record Pairs Side-By-Side.
Remember: Do not bulk label. In this stage of initial training, Tamr Core values quality over quantity. See Working with Tamr Core Machine Learning Models.
When you have a solid set of matching and non-matching pairs identified, have a curator Apply feedback and update results to update Tamr Core. See Curating Record Pairs.
Updated about 1 year ago