World Happiness Report

This example demonstrates how to build and evaluate multiple regression pipelines using the World Happiness Report dataset. The goal is to predict the happiness score of a country based on various features.

Data Source

• Input Topic: World_Happiness_Report
• Target Feature: score (the happiness score to predict)
• Dropped Feature: overall_rank (not used for prediction)

Preprocessing

• Feature Selection: Numeric (int, float) and string (str) features are selected separately.
• Scaling: Numeric features are standardized using StandardScaler.
• Encoding: String features are one-hot encoded with OneHotEncoder.

Algorithms

Three regression algorithms are used:

• LinearRegression: A simple linear model with a specified learning rate for the intercept.
• HoeffdingAdaptiveTreeRegressor: An adaptive tree-based regressor suitable for streaming data.
• KNNRegressor: A k-nearest neighbors regressor.

Pipelines

Each pipeline uses the same data and preprocessing steps but a different regression algorithm:

• linearRegPipeline: Uses LinearRegression.
• hoeffTreePipeline: Uses HoeffdingAdaptiveTreeRegressor.
• knnRegPipeline: Uses KNNRegressor.

Each pipeline outputs predictions to its own topic and evaluates performance using three metrics:

• MAE (Mean Absolute Error)
• MSE (Mean Squared Error)
• R2 (R-squared Score)

Beaver File Structure

Connector

We start by defining the connector, specifying the Kafka bootstrap servers and security protocol.

connector {
        bootstrap_servers = "localhost:39092"
        security_protocol = "plaintext"
        consumer_group = 'world-happiness'
        auto_offset_reset = "earliest"
}

Models

We define the regression algorithms:

algorithm <LinearRegression> linearReg
    params:
        intercept_lr=0.1

algorithm <HoeffdingAdaptiveTreeRegressor> hoeffTree
    params:
        grace_period=50,
        model_selector_decay=0.3,
        seed=0

algorithm <KNNRegressor> knnReg

Feature Selection and Preprocessing

We select numeric and string features separately and apply the appropriate preprocessors:

composer <SelectType> select
    params:
        (int , float)

composer <SelectType> selectstr
    params:
        str

preprocessor <OneHotEncoder> encoder
preprocessor <StandardScaler> scaler

Metrics

We define the evaluation metrics:

metric <MAE> mae
metric <MSE> mse
metric <R2> r2

Data

We define the data source, drop the overall_rank feature, and specify the target and preprocessors:

data World_Happiness_Report {

    input_topic = "World_Happiness_Report"
    features:
        drop_features = overall_rank
        target_feature = score
    preprocessors = select | scaler + selectstr | encoder

}

Pipelines

We define three pipelines, each using a different regression algorithm:

pipeline linearRegPipeline {
    output_topic = 'linearRegPipeline'
    data = World_Happiness_Report
    algorithm = linearReg
    metrics = mae , mse , r2
}

pipeline hoeffTreePipeline {
    output_topic = 'hoeffTreePipeline'
    data = World_Happiness_Report
    algorithm = hoeffTree
    metrics = mae , mse , r2
}

pipeline knnRegPipeline {
    output_topic = 'knnRegPipeline'
    data = World_Happiness_Report
    algorithm = knnReg
    metrics = mae , mse , r2
}

With this configuration, you can efficiently compare multiple regression approaches on the World Happiness Report dataset and gain insights into which model performs best for predicting happiness scores.

connector {
        bootstrap_servers = "localhost:39092"
        security_protocol = "plaintext"
        consumer_group = 'world-happiness'
        auto_offset_reset = "earliest"
}

algorithm <LinearRegression> linearReg
    params:
        intercept_lr=0.1

algorithm <HoeffdingAdaptiveTreeRegressor> hoeffTree
    params:
        grace_period=50,
        model_selector_decay=0.3,
        seed=0

algorithm <KNNRegressor> knnReg

composer <SelectType> select
    params:
        (int , float)

composer <SelectType> selectstr
    params:
        str


preprocessor <OneHotEncoder> encoder

preprocessor <StandardScaler> scaler

metric <MAE> mae

metric <MSE> mse

metric <R2> r2

data World_Happiness_Report {

    input_topic = "World_Happiness_Report"
    features:
        drop_features = overall_rank
        target_feature = score
    preprocessors = select | scaler + selectstr | encoder

}


pipeline linearRegPipeline {
    output_topic = 'linearRegPipeline'
    data = World_Happiness_Report
    algorithm = linearReg
    metrics = mae , mse , r2
}

pipeline hoeffTreePipeline {
    output_topic = 'hoeffTreePipeline'
    data = World_Happiness_Report
    algorithm = hoeffTree
    metrics = mae , mse , r2
}

pipeline knnRegPipeline {
    output_topic = 'knnRegPipeline'
    data = World_Happiness_Report
    algorithm = knnReg
    metrics = mae , mse , r2
}

A complete figure of Beaver Model can be seen below