This section gives you an overview of DASE components and how to implement them. You will find links to some engine templates for more concrete examples.
DataSource

DataSource reads and selects useful data from the Event Store (data store of the Event Server) and returns TrainingData.
readTraining()

You need to implment readTraining() of PDataSource, where you can use the PEventStore Engine API to read the events and create the TrainingData based on the events.
The following code example reads user "view" and "buy" item events, filters specific type of events for future processing and returns TrainingData accordingly.
1
+Implementing DASE™
TEMPLATES OPEN SOURCE
PredictionIO Docs
Implement DASE
PredictionIO Docs
Apache PredictionIO™ DocumentationWelcome to Apache PredictionIO
Getting StartedA Quick Intro
Installing Apache PredictionIO
Downloading an Engine Template
Deploying Your First Engine
Customizing the Engine
Integrating with Your AppApp Integration Overview
List of SDKsJava & Android SDK
PHP SDK
Python SDK
Ruby SDK
Community Powered SDKs
Deploying an EngineDeploying as a Web Service
Batch Predictions
Monitoring En
 gine
Setting Engine Parameters
Deploying Multiple Engine Variants
Engine Server Plugin
Customizing an EngineLearning DASE
Implement DASE
Troubleshooting Engine Development
Engine Scala APIs
Collecting and Analyzing Data
 Event Server Overview
Collecting Data with REST/SDKs
Events Modeling
Unifying Multichannel Data with Webhooks
Channel
Importing Data in Batch
Using Analytics Tools
Event Server Plugin
Choosing an Algorith
 m(s)Built-in Algorithm Libraries
Switching to Another Algorithm
Combining Multiple Algorithms
Adding Your Own Algorithms
ML Tuning and EvaluationOverview
Hyperparameter Tuning
Evaluation Dashboard
Choosing Evaluation Metrics
Building Evaluation Metrics
System ArchitectureArchitecture Overview
Using Another Data Store
PredictionIO Official TemplatesIntro
RecommendationQuick Start
DASE
Evaluation Explained
How-To
Read Custom Events
Customize Data Preparator
Customize Serving
Train with Implicit Preference
Filter Recommended Items by Blacklist in Query
Batch Persistable
  Evaluator
E-Commerce RecommendationQuick Start
DASE
How-To
Train with Rate Event
Adjust Score
Similar ProductQuick Start
DASE
How-To
Multiple Events and Multiple Algorithms
Returns Item Properties
Train with Rate Event
Get Rid of Events for Users
Recommend Users
ClassificationQuick Start
DASE
How-To
Use Alternative Algorithm
Read Custom Properties
Engine Template GalleryBrowse
Submit your Engine as a Template
Demo Tutorials
 Comics Recommendation Demo
Community Contributed Demo
Text Classification Engine Tutorial
Getting InvolvedContribute Code
Contribute Documentation
Contribute a SDK
Contribute a Webhook
Commu
 nity Projects
Getting HelpFAQs
Support
ResourcesCommand-line Interface
Release Cadence
Developing Engines with IntelliJ IDEA
Upgrade Instructions
Glossary
Apache Software Foundation<
 /span>Apache Homepage
License
Sponsorship
Thanks
Security
Customizing an Engine>
Implement DASE
Implementing DASE
On this page
  DataSource   readTraining() 
  Using PEventStore Engine API 
 
 
  Preparator   prepare() 
 
 
  Algorithm   train() 
  predict() 
  P2LAlgorithm 
  PAlgorithm 
  using LEventStore Engine API in predict() 
 
 
  Serving   serve() 
 
 
 
 
Edit this page
Customizing an Engine>
Implement DASE
Implementing DASE
 This section gives you an overview of DASE components and how to implement them. You will find links to some engine templates for more concrete examples.
DataSource
DataSource reads and selects useful data from the Event Store (data store of the Event Server) and returns TrainingData.
readTraining()
You need to implment readTraining() of PDataSource, where you can use the PEventStore Engine API to read the events and create the TrainingData based on the events.
The following code example reads user "view" and "buy" item events, filters specific type of events for future processing and returns TrainingData accordingly.
1
 2
 3
 4
@@ -51,7 +51,7 @@
   }
 
 }
-
 
 Using PEventStore Engine API
Please see Event Server Overview to understand EventAPI and event modeling.
With PEventStore Engine API, you can easily read different events in DataSource and get the information you need.
For example, let's say you have events like the following:
1
+
 
 Using PEventStore Engine API
Please see Event Server Overview to understand EventAPI and event modeling.
With PEventStore Engine API, you can easily read different events in DataSource and get the information you need.
For example, let's say you have events like the following:
1
 2
 3
 4
@@ -123,7 +123,7 @@
           throw e
       }
     }
-
 
 If you have used special events $set/$unset/$delete setting entity's properties, you can retrieve it with PEventStore.aggregateProperties().
Please see event modeling to understand usage of special $set/$unset/$delete events.
For example, the following code show how you could retrieve properties of the "item" entities:
1
+
 
 If you have used special events $set/$unset/$delete setting entity's properties, you can retrieve it with PEventStore.aggregateProperties().
Please see event modeling to understand usage of special $set/$unset/$delete events.
For example, the following code show how you could retrieve properties of the "item" entities:
1
 2
 3
 4
@@ -168,7 +168,7 @@
       }
 
     }
-
 
 Example:
  DataSource of Similar Product Template
 
 Preparator
Preparator is responsible for pre-processing TrainingData for any necessary feature selection and data processing tasks and generate PreparedData which contains the data the Algorithm needs.
A few example usages of Preparator:
  Feature extraction
 Common pre-processing logic if you have multiple algorithms
 For simple cases, the Preparator may simply pass the same TrainingData as PreparedData for Algorithm.
 
 prepare()
You need to implement the prepare() method of PPrepartor to perform such t
 asks.
Example:
  Preparator of Leading Scoring Template: it pre-processes the TrainingData and generate the feature vectors needed for the algorithm.
 Preparator of Similar Product Template: it simply passes the TrainingData as PreparedData for the algorithm.
 
 Algorithm
The two methods of the Algorithm class are train() and predict():
train()
train() is responsible for training a predictive model. It is called when you run pio train. Apache PredictionIO (incubating) will store this model.
predict()
predict() is responsible for using this model to make prediction. It is called when you send a JSON query to the engine. Note that predict() is called in real time.
Apache PredictionIO (inc
 ubating) supports two types of algorithms:
  P2LAlgorithm: trains a Model which does not contain RDD
 PAlgorithm: trains a Model which contains RDD
 
 P2LAlgorithm
For P2LAlgorithm, the Model is automatically serialized and persisted by Apache PredictionIO (incubating) after training.
Implementing IPersistentModel and IPersistentModelLoader is optional for P2LAlgorithm.
Example:
  Algorithm of Similar Product Template
 
 PAlgorithm
PAlgorithm should be used w
 hen your Model contains RDD. The model produced by PAlgorithm is not persisted by default. To persist the model, you need to do the following:
  The Model class should extend the IPersistentModel trait and implement the save() method for saving the model. The trait IPersistentModel requires a type parameter which is the class type of algorithm parameter.
 Implement a Model factory object which extends the IPersistentModelLoader trait and implement the apply() for loading the model. The trait IPersistentModelLoader requires two type parameters which are the types of algorithm parameter and the model produced by the algorithm.
 
 Example:
  Algorithm of Recommendation Template: it implements PAlgorithm and the IPersistentModel and IPersistentModelLoader.
 Alg
 orithm of Vanilla Template: it walks through example of P2LAlgorithm and PAlgorithm.
 
 using LEventStore Engine API in predict()
You may use LEventStore.findByEntity() to retrieve events of a specific entity. For example, retrieve recent events of the user specified in the query) and use these recent events to make prediction in real time.
For example, the following code reads the recent 10 view events of query.user:
1
+
 
 Example:
  DataSource of Similar Product Template
 
 Preparator
Preparator is responsible for pre-processing TrainingData for any necessary feature selection and data processing tasks and generate PreparedData which contains the data the Algorithm needs.
A few example usages of Preparator:
  Feature extraction
 Common pre-processing logic if you have multiple algorithms
 For simple cases, the Preparator may simply pass the same TrainingData as PreparedData for Algorithm.
 
 prepare()
You need to implement the prepare() method of PPrepartor to perform such tasks.<
 p>Example:
  Preparator of Leading Scoring Template: it pre-processes the TrainingData and generate the feature vectors needed for the algorithm.
 Preparator of Similar Product Template: it simply passes the TrainingData as PreparedData for the algorithm.
 
 Algorithm
The two methods of the Algorithm class are train() and predict():
train()
train() is responsible for training a predictive model. It is called when you run pio train. Apache PredictionIO will store this model.
predict()
predict() is responsible for using this model to make prediction. It is called when you send a JSON query to the engine. Note that predict() is called in real time.
Apache PredictionIO supports two types of algor
 ithms:
  P2LAlgorithm: trains a Model which does not contain RDD
 PAlgorithm: trains a Model which contains RDD
 
 P2LAlgorithm
For P2LAlgorithm, the Model is automatically serialized and persisted by Apache PredictionIO after training.
Implementing IPersistentModel and IPersistentModelLoader is optional for P2LAlgorithm.
Example:
  Algorithm of Similar Product Template
 
 PAlgorithm
PAlgorithm should be used when your Model contains RDD. The model produced by PAlgorithm is not persisted by default. To persist the model, you need to do the following:
  The Model class should extend the IPersistentModel trait and implement the save() method for saving the model. The trait IPersistentModel requires a type parameter which is the class type of algorithm parameter.
 Implement a Model factory object which extends the IPersistentModelLoader trait and implement the apply() for loading the model. The trait IPersistentModelLoader requires two type parameters which are the types of algorithm parameter and the model produced by the algorithm.
 
 Example:
  Algorithm of Recommendation Template: it implements PAlgorithm and the IPersistentModel and IPersistentModelLoader.
 Algorithm of Vanilla Template: it walks through example of P2LAlgori
 thm and PAlgorithm.
 
 using LEventStore Engine API in predict()
You may use LEventStore.findByEntity() to retrieve events of a specific entity. For example, retrieve recent events of the user specified in the query) and use these recent events to make prediction in real time.
For example, the following code reads the recent 10 view events of query.user:
1
 2
 3
 4
@@ -211,7 +211,7 @@
         logger.error(s"Error when read recent events: ${e}")
         throw e
     }
-
 
 Example:
  Algorithm of E-Commerce Recommendation template: LEventStore.findByEntity() is used to retrieve all items seen by the user and filter them from recommendation in predict().
 
 Serving
serve()
You need to implement the serve() method of the class LServing. The serve() method processes predicted result. It is also responsible for combining multiple predicted results into one if you have more than one predictive model.
Example:
  Serving of Similar Product Template: It simply returns the predicted result
 Serving
  of multi-algorithm examples of Similar Product Template: It combines the result of multiple algorithms and return
 
 
Community
Download
Docs
GitHub
Subscribe to User Mailing List
Stackoverflow
Contribute
Contribute
Source Code
Bug Tracker
Subscribe to Development Mailing List
Apache PredictionIO, PredictionIO, Apache, the Apache feather logo, and the Apache PredictionIO project logo are either registered trademarks or trademarks of The Apache Software Foundation in the United States and other countries.
All other marks mentioned may be trademarks or registered trademarks of their respective owners.
Apache PredictionIO is an effort undergoing incubation at The Apache Software Foundation (ASF), sponsored by the Apache Incubator. Incubation is required of all newly accepted projects until a further review indicates that the infrastructure, communications, and decision making process have stabilized in a manner consistent with other successful ASF projects. While incubation status is not necessarily a reflection of the completeness or stability of the code, it does indicate that the project has yet to be fully endorsed by the ASF.
™
Star Fork