Extension attributes are not supported in PMML models
Valid from Pega Version 7.3.1
Models in the Predictive Model Markup Language (PMML) format version 4.3 that contain extension attributes with the x- prefix are not valid. These extension attributes are deprecated; you must use extension elements instead. In addition, if the output type of any output field in the model is set to FLOAT, change it to DOUBLE.
For more information, see PMML 4.3 - General Structure in the Data Mining Group documentation.
The Upload responses action is not supported for adaptive models with customized context
Valid from Pega Version 7.3.1
A default instance of the Adaptive Model rule contains five model identifiers (.pyIssue, .pyGroup, .pyName, .pyDirection, .pyChannel) that are used to partition adaptive models. If you add other identifiers in your Adaptive Model rule instance, you cannot upload responses to this instance with the Upload Responses wizard and the following error is displayed: The Flow Action post-processing activity pzUploadCSVFile failed: Cannot parse csv file.You can still train such adaptive models with data flows.
For more information, see Training adaptive models in bulk with data flows, Model context, and Uploading customer responses.
Decision Analytics portal renamed to Analytics Center
Valid from Pega Version 7.3.1
The Decision Analytics portal is renamed to the Analytics Center portal. In this work area for predictive analytics and text analytics, the business scientist can control the full model life cycle by building predictive models, importing PMML models, building sentiment analysis and text classification models, creating and monitoring adaptive models, and updating any existing models.
For more information, Analytics Center portal.
Enhanced adaptive model reporting
Valid from Pega Version 7.4
The new Model report replaces the Behavior and the Performance overview reports to improve report usability and provide consistent information. You can export your Model reports into PDF or Excel files to view or share them outside the Pega® Platform. The Model report also includes information on the groups of correlated predictors where the best performing predictor from each group is active in the model and other remain inactive; this information helps you understand why predictors are active or inactive.
For more information, see Generating a model report.
Use Kinesis data sets in Pega Decision Management
Valid from Pega Version 7.4
You can create Kinesis data set instances to connect to Amazon Kinesis Data Streams and use this data set in decision management for processing real-time streaming data. Integrating Kinesis data streams into Pega® Platform in the cloud provides a fault-tolerant and scalable solution for processing IT infrastructure log data, application logs, social media, market data feeds, and web clickstream data.
For more information, see Creating a Kinesis data set.
Store and scale the processing of Stream data records on multiple nodes
Valid from Pega Version 7.4
You can configure the Stream service on Pega® Platform to ingest, route, and deliver high volumes of low-latency data such as web clicks, transactions, sensor data, and customer interaction history. You can store streams of records in a fault-tolerant way and process stream records as they occur. Add or remove Stream nodes to increase or decrease the use of the Stream service and optimize data processing.
For more information, see Stream service overview.
Decisioning services now use default node classification
Valid from Pega Version 7.4
Decisioning services have been integrated with default node classification on Pega® Platform to provide a unified way of creating and initializing services. As a result of the integration, the Data Flow service has been divided into Batch and Real Time services to better handle different types of data flow runs. You can now specify separate subsets of Data Flow nodes for batch data flow runs and real-time data flow runs to divide the workload between these two subsets.
For more information, see Node classification, Data Flows landing page, and Services landing page.
Train machine learning models for extracting named entities and detecting intents
Valid from Pega Version 7.4
Data scientists can train machine learning-based text extraction and intent detection models by using the Analytics Center. With text extraction, you can train a Conditional Random Fields (CRF) model to detect whether the content contains specified entity types such as person names, company and organization names, locations, dates and times, percentages, and monetary amounts. For intent detection, you can train a maximum entropy model to understand user intentions expressed in written content. With these two new capabilities, you can quickly react to customer queries and comments by taking appropriate action against the information that you extracted.
For more information, see Creating machine learning-based text extraction models and Creating machine learning-based intent analysis models.
Label changes for text analytic models
Valid from Pega Version 7.4
The classification analysis label has changed to topic detection and the entity extraction label has changed to text extraction. Also, the sentiment analysis, topic detection, and intent detection labels are now located under Text Categorization in the Analytics Center. These name changes reflect industry standards and provide a clearer distinction between different types of text analytics models in Pega® Platform.
For more information, see Text analytics models.
Automatically detect the most important themes in text
Valid from Pega Version 7.4
Text analyzers can now automatically recognize the most important concepts that are expressed in text and mark such concepts as entities of type auto_tags. You can use auto_tags to group similar content by its themes and reduce the dimensionality of text to the most important features.
For more information, see Text extraction analysis.