生存回归 (AftSurvivalRegression)

Java 类名:com.alibaba.alink.pipeline.regression.AftSurvivalRegression

Python 类名:AftSurvivalRegression

功能介绍

在生存分析领域,加速失效时间模型(accelerated failure time model,AFT 模型)可以作为比例风险模型的替代模型。生存回归组件支持稀疏、稠密两种数据格式。

算法原理

AFT模型将线性回归模型的建模方法引人到生存分析的领域, 将生存时间的对数作为反应变量,研究多协变量与对数生存时间之间的回归关系,在形式上,模型与一般的线性回归模型相似。对回归系数的解释也与一般的线性回归模型相似,较之Cox模型, AFT模型对分析结果的解释更加简单、直观且易于理解,并且可以预测个体的生存时间。

算法使用

生存回归分析是研究特定事件的发生与时间的关系的回归。这里特定事件可以是:病人死亡、病人康复、用户流失、商品下架等。

  • 备注 :该组件训练的时候 FeatureCols 和 VectorCol 是两个互斥参数,只能有一个参数来描述算法的输入特征。

文献或出处

[1] Wei, Lee-Jen. “The accelerated failure time model: a useful alternative to the Cox regression model in survival analysis.” Statistics in medicine 11.14‐15 (1992): 1871-1879.

[2] https://spark.apache.org/docs/latest/ml-classification-regression.html#survival-regression

名称 中文名称 描述 类型 是否必须? 取值范围 默认值
censorCol 生存列名 生存列名 String
labelCol 标签列名 输入表中的标签列名 String
predictionCol 预测结果列名 预测结果列名 String
epsilon 收敛阈值 迭代方法的终止判断阈值,默认值为 1.0e-6 Double x >= 0.0 1.0E-6
featureCols 特征列名数组 特征列名数组,默认全选 String[] null
l1 L1 正则化系数 L1 正则化系数,默认为0。 Double x >= 0.0 0.0
l2 L2 正则化系数 L2 正则化系数,默认为0。 Double x >= 0.0 0.0
maxIter 最大迭代步数 最大迭代步数,默认为 100 Integer x >= 1 100
modelFilePath 模型的文件路径 模型的文件路径 String null
overwriteSink 是否覆写已有数据 是否覆写已有数据 Boolean false
predictionDetailCol 预测详细信息列名 预测详细信息列名 String
quantileProbabilities 分位数概率数组 分位数概率数组 double[] [0.01,0.05,0.1,0.25,0.5,0.75,0.9,0.95,0.99]
reservedCols 算法保留列名 算法保留列 String[] null
vectorCol 向量列名 向量列对应的列名,默认值是null String null
withIntercept 是否有常数项 是否有常数项,默认true Boolean true
numThreads 组件多线程线程个数 组件多线程线程个数 Integer 1
modelStreamFilePath 模型流的文件路径 模型流的文件路径 String null
modelStreamScanInterval 扫描模型路径的时间间隔 描模型路径的时间间隔,单位秒 Integer 10
modelStreamStartTime 模型流的起始时间 模型流的起始时间。默认从当前时刻开始读。使用yyyy-mm-dd hh:mm:ss.fffffffff格式,详见Timestamp.valueOf(String s) String null

代码示例

Python 代码

from pyalink.alink import *

import pandas as pd

useLocalEnv(1)

df = pd.DataFrame([
    [1.218, 1.0, "1.560,-0.605"],
    [2.949, 0.0, "0.346,2.158"],
    [3.627, 0.0, "1.380,0.231"],
    [0.273, 1.0, "0.520,1.151"],
    [4.199, 0.0, "0.795,-0.226"]])

data = BatchOperator.fromDataframe(df, schemaStr="label double, censor double, features string")

reg = AftSurvivalRegression()\
            .setVectorCol("features")\
            .setLabelCol("label")\
            .setCensorCol("censor")\
            .setPredictionCol("result")

pipeline = Pipeline().add(reg)
model = pipeline.fit(data)

model.save().lazyPrint(10)
model.transform(data).print()

Java 代码

import org.apache.flink.types.Row;

import com.alibaba.alink.operator.batch.BatchOperator;
import com.alibaba.alink.operator.batch.source.MemSourceBatchOp;
import com.alibaba.alink.pipeline.Pipeline;
import com.alibaba.alink.pipeline.PipelineModel;
import com.alibaba.alink.pipeline.regression.AftSurvivalRegression;
import org.junit.Test;

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

public class AftSurvivalRegressionTest {
	@Test
	public void testAftSurvivalRegression() throws Exception {
		List <Row> df = Arrays.asList(
			Row.of(1.218, 1.0, "1.560,-0.605"),
			Row.of(2.949, 0.0, "0.346,2.158"),
			Row.of(3.627, 0.0, "1.380,0.231"),
			Row.of(0.273, 1.0, "0.520,1.151")
		);
		BatchOperator <?> data = new MemSourceBatchOp(df, "label double, censor double, features string");
		AftSurvivalRegression reg = new AftSurvivalRegression()
			.setVectorCol("features")
			.setLabelCol("label")
			.setCensorCol("censor")
			.setPredictionCol("result");
		Pipeline pipeline = new Pipeline().add(reg);
		PipelineModel model = pipeline.fit(data);
		model.save().lazyPrint(10);
		model.transform(data).print();
	}
}

运行结果

模型

id p0 p1 p2
-1 {“stages”:“[{"identifier":null,"params":null,"schemaIndices":[1,0,2],"colNames":null,"parent":-1},{"identifier":"com.alibaba.alink.pipeline.regression.AftSurvivalRegressionModel","params":{"params":{"vectorCol":"\"features\"","labelCol":"\"label\"","censorCol":"\"censor\"","predictionCol":"\"result\""}},"schemaIndices":[1,0,2],"colNames":["model_id","model_info","label_value"],"parent":0}]”} null null
1 {“hasInterceptItem”:“true”,“vectorCol”:“"features"”,“modelName”:“"AFTSurvivalRegTrainBatchOp"”,“labelCol”:“"label"”,“linearModelType”:“"AFT"”,“vectorSize”:“3”} 0 null
1 {“featureColNames”:null,“featureColTypes”:null,“coefVector”:{“data”:[-29.487324590178716,24.42773010344541,13.44725070039797,-1.3679961023031253]},“coefVectors”:null,“convergenceInfo”:[1.5843984652595493,3.6032723097911044,0.4,1.4794299745122195,0.9580954096270979,1.6,1.3777797119903465,0.7050802052575507,1.6,1.3399286821587995,0.3682693394041936,1.6,1.312648708021441,0.24739884143507995,4.0,1.2685626011340911,0.18750659133206055,4.0,1.253583736945237,0.14860925947925266,4.0,1.2281061305710799,0.14586073980515185,4.0,1.0942468743404496,0.18594588792948594,4.0,0.8350708072737613,0.3504767418363587,4.0,0.8350708072737618,0.5905879812330285,0.25,0.5762249843357561,0.5905879812330285,0.25,0.5161276782605526,0.7008299684632876,0.25,0.3872690319921853,0.6482128538375713,1.0,0.3872690319921861,0.6448719615922804,0.0625,0.3786668764217095,0.6448719615922804,0.0625,0.3354192551580446,3.6845873037311816,0.25,0.26183684259256024,3.141816288006177,1.0,-0.07097230167077419,2.6843478562459735,1.0]} 1048576 null

结果

label censor features result
1.2180 1.0000 1.560,-0.605 1.6231
2.9490 0.0000 0.346,2.158 2933.1642
3.6270 0.0000 1.380,0.231 1524.2502
0.2730 1.0000 0.520,1.151 0.2706