从 Apache Flink ML 中提取 LinearSVC 模型系数和截距

from pyflink.common import Types
from pyflink.table import (
    DataTypes,
    StreamTableEnvironment,
    TableDescriptor,
    Schema,
)
from pyflink.ml.linalg import Vectors, DenseVector
from pyflink.ml.classification.linear_svc import LinearSVC
from pyflink.ml.common import Params
import os
import tempfile

# 创建一个临时目录用于存储模型数据
tmp_dir = tempfile.mkdtemp()
model_path = os.path.join(tmp_dir, "linear_svc_model")

# 创建一个 TableEnvironment
t_env = StreamTableEnvironment.create(
    environment_settings=StreamTableEnvironment.DEFAULT_STREAMING
)

# 定义输入数据模式
input_schema = Schema.new_builder() \
    .add_column("features", DataTypes.ARRAY(DataTypes.DOUBLE())) \
    .add_column("label", DataTypes.DOUBLE()) \
    .build()

# 创建一个 TableDescriptor，用于定义输入表
input_data = t_env.from_descriptor(
    TableDescriptor.for_connector("datagen")
        .schema(input_schema)
        .option("number-of-rows", "10")
        .build()
)

# 创建 LinearSVC 模型
linear_svc = LinearSVC() \
    .set_features_col("features") \
    .set_label_col("label") \
    .set_prediction_col("prediction")

# 训练模型
model = linear_svc.fit(input_data)

# 保存模型
model.save(model_path)

# 加载模型
loaded_model = LinearSVC.load(model_path)

# 获取模型数据
model_data = loaded_model.get_model_data()[0]

# 提取系数和截距
coefficients = model_data.coefficients
intercept = model_data.intercept

print("Coefficients:", coefficients)
print("Intercept:", intercept)

import org.apache.flink.ml.classification.LinearSVC;
import org.apache.flink.ml.classification.LinearSVCModel;
import org.apache.flink.ml.linalg.DenseVector;
import org.apache.flink.table.api.Table;
import org.apache.flink.table.api.bridge.java.StreamTableEnvironment;
import org.apache.flink.types.Row;
import org.apache.flink.api.java.tuple.Tuple2;
import org.apache.flink.api.java.tuple.Tuple3;

import java.util.Arrays;
import java.util.List;

// 示例数据 (features, label)
List<Tuple2<DenseVector, Double>> data = Arrays.asList(
        Tuple2.of(new DenseVector(new double[]{1.0, 2.0}), 0.0),
        Tuple2.of(new DenseVector(new double[]{3.0, 4.0}), 1.0)
);

// 将数据转换为 Table
StreamTableEnvironment tEnv = StreamTableEnvironment.create(env);
Table table = tEnv.fromDataStream(env.fromCollection(data).map(x -> Row.of(x.f0, x.f1)));

// 注册表
tEnv.createTemporaryView("inputTable", table, "features, label");

// 创建 LinearSVC 模型
LinearSVC linearSVC = new LinearSVC()
        .setFeaturesCol("features")
        .setLabelCol("label")
        .setPredictionCol("prediction");

// 训练模型
LinearSVCModel model = linearSVC.fit(table);

// 获取模型数据
List<Tuple3<Double, DenseVector, Double>> modelData = model.getModelData().executeAndCollect();

// 提取系数和截距
DenseVector coefficients = modelData.get(0).f1;
double intercept = modelData.get(0).f0;

System.out.println("Coefficients: " + coefficients);
System.out.println("Intercept: " + intercept);