程序包 org.apache.flink.ml.feature.kbinsdiscretizer

类 KBinsDiscretizer

java.lang.Object

org.apache.flink.ml.feature.kbinsdiscretizer.KBinsDiscretizer

所有已实现的接口:

Serializable, org.apache.flink.ml.api.Estimator<KBinsDiscretizer,KBinsDiscretizerModel>, org.apache.flink.ml.api.Stage<KBinsDiscretizer>, org.apache.flink.ml.common.param.HasInputCol<KBinsDiscretizer>, org.apache.flink.ml.common.param.HasOutputCol<KBinsDiscretizer>, KBinsDiscretizerModelParams<KBinsDiscretizer>, KBinsDiscretizerParams<KBinsDiscretizer>, org.apache.flink.ml.param.WithParams<KBinsDiscretizer>

public class KBinsDiscretizer
extends Object
implements org.apache.flink.ml.api.Estimator<KBinsDiscretizer,KBinsDiscretizerModel>, KBinsDiscretizerParams<KBinsDiscretizer>

An Estimator which implements discretization (also known as quantization or binning) to transform continuous features into discrete ones. The output values are in [0, numBins).

KBinsDiscretizer implements three different binning strategies, and it can be set by KBinsDiscretizerParams.STRATEGY. If the strategy is set as KBinsDiscretizerParams.KMEANS or KBinsDiscretizerParams.QUANTILE, users should further set KBinsDiscretizerParams.SUB_SAMPLES for better performance.

There are several corner cases for different inputs as listed below:

• When the input values of one column are all the same, then they should be mapped to the same bin (i.e., the zero-th bin). Thus the corresponding bin edges are `{Double.NEGATIVE_INFINITY, Double.POSITIVE_INFINITY}`.
• When the number of distinct values of one column is less than the specified number of bins and the KBinsDiscretizerParams.STRATEGY is set as KBinsDiscretizerParams.KMEANS, we switch to KBinsDiscretizerParams.UNIFORM.
• When the width of one output bin is zero, i.e., the left edge equals to the right edge of the bin, we replace the right edge as the average value of its two neighbors. One exception is that the last two edges are the same --- in this case, the left edge is updated as the average of its two neighbors. For example, the bin edges {0, 1, 1, 2, 2} are transformed into {0, 1, 1.5, 1.75, 2}.

另请参阅:

序列化表格

字段概要

从接口继承的字段 org.apache.flink.ml.common.param.HasInputCol

INPUT_COL

从接口继承的字段 org.apache.flink.ml.common.param.HasOutputCol

OUTPUT_COL

从接口继承的字段 org.apache.flink.ml.feature.kbinsdiscretizer.KBinsDiscretizerParams

KMEANS, NUM_BINS, QUANTILE, STRATEGY, SUB_SAMPLES, UNIFORM

构造器概要

构造器

构造器	说明
KBinsDiscretizer()

方法概要

修饰符和类型	方法	说明
KBinsDiscretizerModel	fit(org.apache.flink.table.api.Table... inputs)
Map<org.apache.flink.ml.param.Param<?>,Object>	getParamMap()
static KBinsDiscretizer	load(org.apache.flink.table.api.bridge.java.StreamTableEnvironment tEnv, String path)
void	save(String path)

从类继承的方法 java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

从接口继承的方法 org.apache.flink.ml.common.param.HasInputCol

getInputCol, setInputCol

从接口继承的方法 org.apache.flink.ml.common.param.HasOutputCol

getOutputCol, setOutputCol

从接口继承的方法 org.apache.flink.ml.feature.kbinsdiscretizer.KBinsDiscretizerParams

getNumBins, getStrategy, getSubSamples, setNumBins, setStrategy, setSubSamples

从接口继承的方法 org.apache.flink.ml.param.WithParams

get, getParam, set

构造器详细资料

KBinsDiscretizer

public KBinsDiscretizer()

方法详细资料

fit

public KBinsDiscretizerModel fit(org.apache.flink.table.api.Table... inputs)

指定者:

fit 在接口中 org.apache.flink.ml.api.Estimator<KBinsDiscretizer,KBinsDiscretizerModel>

getParamMap

public Map<org.apache.flink.ml.param.Param<?>,Object> getParamMap()

指定者:

getParamMap 在接口中 org.apache.flink.ml.param.WithParams<KBinsDiscretizer>

save

public void save(String path)
          throws IOException

指定者:

save 在接口中 org.apache.flink.ml.api.Stage<KBinsDiscretizer>

抛出:

IOException

load

public static KBinsDiscretizer load(org.apache.flink.table.api.bridge.java.StreamTableEnvironment tEnv,
                                    String path)
                             throws IOException

抛出:

IOException