Introduction

The Machine Learning Platform in TIBCO Patterns uses a machine learning model to make “Yes” or “No” decisions for problems that can be characterized by a particular set of features. In the context of the Machine Learning Platform, a feature is any characteristic of a record matching problem where the match of two data items can be quantified as a real value in the range 0.0 - 1.0.

Where :

•

The value 0.0 represents the “most false” condition for the feature.

•

The value 1.0 represents the “most true” condition for the feature.

•

The numbers in between these values represent proportional degrees of "true" and "false". A larger value is always associated with a more positive human judgment for the feature, or at least an unchanged judgment, but cannot be associated with a decreased judgment.

Traditionally these decisions are made based on a set of manually created rules. The rule sets needed to achieve good results are often large and complex. It can be difficult to understand all of the consequences of changing or adding rules to the rule set. Using a Learn model avoids these problems. A Learn model needs to be trained in order to establish relationships between pairs of records with “True” or “False” labels. The trainer of this model can be anybody who can judge the relationships from the given examples.

The trained model can be used to predict the “True” or “False” labels of novel examples. This eliminates the need for creating an explicit rule set.

You can use Learn User Interface application to conveniently accomplish all the steps of training and evaluating Learn Models in TIBCO Patterns. The models are trained to predict whether any two records in a data table match or do not match. The Learn UI application guides you through preparing the data table, defining features, automatically finding useful record pairs for model training, labeling the record pairs and ensuring consistent labels, performing model training and evaluation, and augmenting the training data to improve model predictions.