bayesianRegression (Stream Processor)

This function predicts using a Bayesian linear regression model. Bayesian linear regression allows determining the uncertainty of each prediction by estimating the full-predictive distribution

Syntax

streamingml:bayesianRegression(<STRING> model.name, <DOUBLE|FLOAT|INT|LONG> model.feature, <DOUBLE|FLOAT|INT|LONG> ...)
streamingml:bayesianRegression(<STRING> model.name, <INT> prediction.samples, <DOUBLE|FLOAT|INT|LONG> model.feature, <DOUBLE|FLOAT|INT|LONG> ...)

Query Parameters

Name	Description	Default Value	Possible Data Types	Optional	Dynamic
model.name	The name of the model to be used.		STRING	No	No
prediction.samples	The number of samples to be drawn to estimate the prediction.	1000	INT	Yes	No
model.feature	The features of the model that need to be attributes of the stream.		DOUBLE FLOAT INT LONG	No	Yes

Extra Return Attributes

Name	Description	Possible Types
prediction	The predicted value (double).	DOUBLE
confidence	Inverse of the standard deviation of the predictive distribution.	DOUBLE

Example 1

CREATE STREAM StreamA (attribute_0 double, attribute_1 double, attribute_2 double, attribute_3 double);
CREATE STREAM PredictionStream (attribute_0 double, attribute_1 double, attribute_2 double, attribute_3 double);

CREATE SINK STREAM OutputStream (prediction double, confidence double, attribute_0 double, attribute_1 double, attribute_2 double, attribute_3 double);

@info(name = 'predictAttributes')
INSERT INTO OutputStream
SELECT prediction, confidence, attribute_0, attribute_1, attribute_2, attribute_3
FROM PredictionStream#streamingml:bayesianRegression('model1', attribute_0, attribute_1, attribute_2, attribute_3);

This query uses a Bayesian regression model named model1 to predict the label of the feature vector represented by attribute_0, attribute_1, attribute_2, and attribute_3. The StreamA contains the input features, and the PredictionStream is created with the same features.

A sink stream, OutputStream, is defined to store the predicted values, prediction confidence (std of predictive distribution), and the original features from the PredictionStream.

The predictAttributes query processes events from the PredictionStream, using the Bayesian regression model 'model1' to make predictions based on the input features. The predicted values, along with their confidences and original features, are inserted into the OutputStream sink stream. As a result, the OutputStream stream is defined as follows: (prediction double, confidence double, attribute_0 double, attribute_1 double, attribute_2 double, attribute_3 double).

Example 2

CREATE STREAM HouseDataStream (square_feet double, num_bedrooms double, num_bathrooms double, age double, price double);
CREATE STREAM HousePredictionStream (square_feet double, num_bedrooms double, num_bathrooms double, age double);

CREATE SINK STREAM HousePricePredictions (prediction double, confidence double, square_feet double, num_bedrooms double, num_bathrooms double, age double);

@info(name = 'trainHousePriceModel')
INSERT INTO OutputStream2
SELECT *
FROM HouseDataStream#streamingml:updateBayesianRegression('housePriceModel', price, square_feet, num_bedrooms, num_bathrooms, age);

@info(name = 'predictHousePrices')
INSERT INTO HousePricePredictions
SELECT prediction, confidence, square_feet, num_bedrooms, num_bathrooms, age
FROM HousePredictionStream#streamingml:bayesianRegression('housePriceModel', square_feet, num_bedrooms, num_bathrooms, age);

In this example, two input streams are created: HouseDataStream for model training and HousePredictionStream for making predictions. The HouseDataStream contains the square footage, number of bedrooms, number of bathrooms, age of the house, and the actual price. The HousePredictionStream contains the same features but without the actual price.

A sink stream, HousePricePredictions, is defined to store the predicted house prices, prediction confidence, and the original features from the HousePredictionStream.

The trainHousePriceModel query processes events from the HouseDataStream, and builds or updates the Bayesian linear regression model 'housePriceModel' using the input features and the actual house prices.

The predictHousePrices query processes events from the HousePredictionStream, using the trained 'housePriceModel' to predict house prices based on the input features. The predictions, along with their confidences and original features, are inserted into the HousePricePredictions sink stream.