2016-09-08 63 views
1

我试图在PySpark中使用我在Kaggle上找到的住房数据集做一个非常简单的LinearRegression。有很多列,但为了尽可能简化(实际上),我只保留了两列(在开始所有列之后),仍然没有运气让模型得到训练。这是该数据帧的样子之前通过回归步打算:找不到Spark LinearRegression的原因错误

2016-09-07 17:12:08,804 root INFO [Row(price=78000.0, sqft_living=780.0, sqft_lot=16344.0, features=DenseVector([780.0, 16344.0])), Row(price=80000.0, sqft_living=430.0, sqft_lot=5050.0, features=DenseVector([430.0, 5050.0])), Row(price=81000.0, sqft_living=730.0, sqft_lot=9975.0, features=DenseVector([730.0, 9975.0])), Row(price=82000.0, sqft_living=860.0, sqft_lot=10426.0, features=DenseVector([860.0, 10426.0])), Row(price=84000.0, sqft_living=700.0, sqft_lot=20130.0, features=DenseVector([700.0, 20130.0])), Row(price=85000.0, sqft_living=830.0, sqft_lot=9000.0, features=DenseVector([830.0, 9000.0])), Row(price=85000.0, sqft_living=910.0, sqft_lot=9753.0, features=DenseVector([910.0, 9753.0])), Row(price=86500.0, sqft_living=840.0, sqft_lot=9480.0, features=DenseVector([840.0, 9480.0])), Row(price=89000.0, sqft_living=900.0, sqft_lot=4750.0, features=DenseVector([900.0, 4750.0])), Row(price=89950.0, sqft_living=570.0, sqft_lot=4080.0, features=DenseVector([570.0, 4080.0]))] 

我用下面的代码来训练模型:

standard_scaler = StandardScaler(inputCol='features', 
            outputCol='scaled') 
    lr = LinearRegression(featuresCol=standard_scaler.getOutputCol(), labelCol='price', weightCol=None, 
          maxIter=100, tol=1e-4) 
    pipeline = Pipeline(stages=[standard_scaler, lr]) 
    grid = (ParamGridBuilder() 
      .baseOn({lr.labelCol: 'price'}) 
      .addGrid(lr.regParam, [0.1, 1.0]) 
      .addGrid(lr.elasticNetParam, elastic_net_params or [0.0, 1.0]) 
      .build()) 
    ev = RegressionEvaluator(metricName="rmse", labelCol='price') 
    cv = CrossValidator(estimator=pipeline, 
         estimatorParamMaps=grid, 
         evaluator=ev, 
         numFolds=5) 
    model = cv.fit(data).bestModel 

我得到的错误是:

2016-09-07 17:12:08,805 root INFO Training regression model... 
2016-09-07 17:12:09,530 root ERROR An error occurred while calling o60.fit. 
: java.lang.NullPointerException 
    at org.apache.spark.ml.regression.LinearRegression.train(LinearRegression.scala:164) 
    at org.apache.spark.ml.regression.LinearRegression.train(LinearRegression.scala:70) 
    at org.apache.spark.ml.Predictor.fit(Predictor.scala:90) 
    at org.apache.spark.ml.Predictor.fit(Predictor.scala:71) 
    at sun.reflect.NativeMethodAccessorImpl.invoke0(Native Method) 
    at sun.reflect.NativeMethodAccessorImpl.invoke(NativeMethodAccessorImpl.java:62) 
    at sun.reflect.DelegatingMethodAccessorImpl.invoke(DelegatingMethodAccessorImpl.java:43) 
    at java.lang.reflect.Method.invoke(Method.java:498) 
    at py4j.reflection.MethodInvoker.invoke(MethodInvoker.java:237) 
    at py4j.reflection.ReflectionEngine.invoke(ReflectionEngine.java:357) 
    at py4j.Gateway.invoke(Gateway.java:280) 
    at py4j.commands.AbstractCommand.invokeMethod(AbstractCommand.java:128) 
    at py4j.commands.CallCommand.execute(CallCommand.java:79) 
    at py4j.GatewayConnection.run(GatewayConnection.java:211) 
    at java.lang.Thread.run(Thread.java:745) 

有什么想法?

回答

1

在这种情况下,您不能使用Pipeline。当你调用pipeline.fit将它转换为(大约)

standard_scaler_model = standard_scaler.fit(dataframe) 
lr_model = lr.fit(dataframe) 

但你确实需要

standard_scaler_model = standard_scaler.fit(dataframe) 
dataframe = standard_scaler_model.transform(dataframe) 
lr_model = lr.fit(dataframe) 

这个错误是因为你的lr.fit找不到输出(即转换的结果),你StandardScaler的模型。

+0

这里的错误不是由StandardScaler造成的。这对我来说很好(显然你的体验不一样)。该错误原来是“权重”列。当我试图指定'weightCol = None'时,对我造成错误。我通过创建一个1.0的weightCol作为重量来固定它(必须是浮点!)。 –