TIBCO Patterns Machine Learning Platform

TIBCO Patterns Machine Learning Platform provides machine learning models and associated supervised learning algorithms that are used for classification.

A set of training examples is provided where each example is marked as belonging to one or the other of two categories (True and False). The training algorithm builds a TIBCO Patterns Learn Model that assigns any new example to one of these categories, making it a binary classifier.

The TIBCO Patterns Machine Learning Platform is commonly used in matching and deduplication of records in a table to predict whether any two given records in a table match (represent the same entity) despite the differences in their field values. The file containing the trained Learn model is loaded to a TIBCO Patterns server and is used to provide more accurate record matches using the same query that was used to train the model. A trained model is associated with the query used to train the model and cannot be used with a different query.

TIBCO Patterns Machine Learning Platform incorporates a machine learning model trained to evaluate instances of problems defined by a particular set of features. In the context of TIBCO Patterns Machine Learning Platform, a feature is any characteristic of a problem that can be expressed as a real value between 0.0 and 1.0:

•

The score 0.0 represents the "most false" condition for the feature.

•

The score 1.0 represents the "most true" condition for the feature.

•

A score between 0.0 and 1.0 represents proportional degrees of "true" and "false." A larger score is always associated with a more positive human judgment for the feature, or at least an unchanged judgment, but cannot be associated with a more negative judgment. That is, a decrease in the score of one feature cannot lead to an increase in the likelihood of the example being considered a true example.

For example, any score output by a query in TIBCO Patterns can serve as an input feature for TIBCO Patterns Learn model. However, the features for a Learn model are not limited to such scores. They can be real values obtained from other sources.

A machine learning model is a data component separate from the TIBCO Patterns server. The model can be created using the TIBCO Patterns Learn UI, programmatically using the TIBCO Patterns Learn API library, or TIBCO can create and tune the model for you as a contract service. For more information about creating a model, see TIBCO Patterns Learn UI User's Guide and the Learn API documentation.

Multiple machine learning models can be used in a single instance of TIBCO Patterns server.

Each model represents the domain intelligence needed to make positive or negative evaluation of feature vectors. A feature vector represents an instance of the classification problem and has one feature value for each feature.