Package com.tibco.patterns.learn.api.project
The main package to support Patterns Learn projects and automated model training.
ProjectFiles class manages all files in the project directory. It creates or loads
one project.xml and a number of model.xml files, and returns objects of ProjectSettings
and ModelSettings for the respective XML files. All model training functionality for
a specific ModelSettings object is implemented by ModelTrainer class.
The package also contains a number of classes that support individual data structures required for Patterns Learn projects, such as DataPartition, DatasetResult, RecPair and RecPairMap.
-
Class Summary Class Description AddPairsResult Contains sets of pairs that were added to each dataset, and collections of pairs that were not added to the project for various reasons.DatasetResult Stores model training results for one dataset.FieldStats Stores statistics for one table field.ModelSettings Stores all settings for a single Patterns Learn model.xml file.ModelTrainConfig Defines configuration for creating and training a Learn model.ModelTrainer Functionality to communicate with Patterns server, load data table, calculate feature scores, train and evaluate model.ProjectFiles Manages all files for one Patterns Learn project.ProjectSettings Stores all settings for a single Patterns Learn project.xml file.RecPair Stores information about one record pair, including field values, feature scores and label.RecPair.ConfScoreComparator Provides ordering by a combination (weighted sum) of the prediction confidence (ascending) and the distance of the model score from the 0.5 threshold (ascending).RecPairMap Manages datasets in the XML structure in sync with a map of all record pairs that are stored in these datasets.ValueFreq Stores one field value and its frequency. -
Enum Summary Enum Description DataPartition Defines the type of a dataset: Training, Validation, Test, Reserve, Low Confidence, False Subsets.RecPair.Label Defines a label that can be assigned to a record pair: True, False or Unsure.