Step-By-Step Guide

Instructions on how to install and use Zingg

Step 1: Install

Installation instructions for docker, as well as GitHub release, are here. If you need to build from the sources or compile for a different flavor of Spark, check compiling.

Step 2: Plan for Hardware

Decide your hardware based on the performance numbers.

Step 3: Build the config for your data

Zingg needs a configuration file that defines the data and what kind of matching is needed. You can create the configuration file by following the instructions here.

Step 4: Create the training data

Zingg builds a new set of models(blocking and similarity) for every new schema definition(columns and match types). This means running the findTrainingData and label phases multiple times to build the training dataset from which Zingg will learn. You can read more here.

Step 5: Build and save the model

The training data in Step 4 above is used to train Zingg and build and save the models. This is done by running the train phase. Read more here.

Step 6: Voila, let's match!

It's now time to apply the model to our data. This is done by running the match or the link phases depending on whether you are matching within a single source or linking multiple sources respectively. You can read more about matching and linking.

As long as your input columns and the field types are not changing, the same model should work and you do not need to build a new model. If you change the match type, you can continue to use the training data and add more labeled pairs on top of it.