经过进一步学习，发现flink-ml-lib这个包是基于flink-ml-api这个包做的一个高层封装。
下面，就对这个包的内容进行深入的分析。

首先这个包里分为四个模块，分别对应本文的四个一级标题。

common

• linalg 主要与线性代数有关
• mapper 一些映射，现在还不知道干什么的
• model 一些model source，作用未知
• statistics 统计学算法，里边只有一个多元高斯分布
• utils 常用工具类
• MLEnvironment 作用未知
• MLEnvironmentFactory 作用未知

linalg

• BLAS
  A utility class that provides BLAS routines over matrices and vectors.
  BLAS（Basic Linear Algebra Subprograms）即基础线性代数子程序库，里面拥有大量已经编写好的关于线性代数运算的程序。

• DenseMatrix
  DenseMatrix stores dense matrix data and provides some methods to operate on the matrix it represents.

• DenseVector
  A dense vector represented by a values array.

• MatVecOp 工具类
  A utility class that provides operations over {@link DenseVector}, {@link SparseVector} and {@link DenseMatrix}.

• SparseVector
  A sparse vector represented by an indices array and a values array.

• Vector 工具类，包含关于DenseVector和SparseVector的常用方法
  The Vector class defines some common methods for both DenseVector and SparseVector.

• VectorIterator 遍历Vector使用的工具类
  An iterator over the elements of a vector.

• VectorUtil Vector和它的子类的工具类
  Utility class for the operations on {@link Vector} and its subclasses.

mapper

• Mapper
  Abstract class for mappers. A mapper takes one row as input and transform it into another row.
• MapperAdapter
  A class that helps adapt a {@link Mapper} to a {@link MapFunction} so that the mapper can run in Flink.
• ModelMapper
  An abstract class for {@link Mapper Mappers} with a model.
• ModelMapperAdapter
  A class that adapts a {@link ModelMapper} to a Flink {@link RichMapFunction} so the model can be loaded in a Flink job.
  This adapter class hold the target {@link ModelMapper} and it’s {@link ModelSource}.
  Upon open(), it will load model rows from {@link ModelSource} into {@link ModelMapper}.

model

• BroadcastVariableModelSource
  A {@link ModelSource} implementation that reads the model from the broadcast variable.

• ModelSource
  An interface that load the model from different sources. E.g. broadcast variables, list of rows, etc.

• RowsModelSource
  A {@link ModelSource} implementation that reads the model from the memory.

statistics - basicstatistic

• MultivariateGaussian 多元高斯分布
  This class provides basic functionality for a Multivariate Gaussian (Normal) Distribution.

utils

• DataSetConversionUtil
  Provide functions of conversions between DataSet and Table.
• DataStreamConversionUtil
  Provide functions of conversions between DataStream and Table.
• OutputColsHelper

/**
 * Utils for merging input data with output data.
 *
 * <p>Input:
 * 1) Schema of input data being predicted or transformed.
 * 2) Output column names of the prediction/transformation operator.
 * 3) Output column types of the prediction/transformation operator.
 * 4) Reserved column names, which is a subset of input data's column names that we want to preserve.
 *
 * <p>Output:
 * 1）The result data schema. The result data is a combination of the preserved columns and the operator's
 * output columns.
 *
 * <p>Several rules are followed:
 * <ul>
 * <li>If reserved columns are not given, then all columns of input data is reserved.
 * <li>The reserved columns are arranged ahead of the operator's output columns in the final output.
 * <li>If some of the reserved column names overlap with those of operator's output columns, then the operator's
 * output columns override the conflicting reserved columns.
 * <li>The reserved columns in the result table preserve their orders as in the input table.
 * </ul>
 *
 * <p>For example, if we have input data schema of ["id":INT, "f1":FLOAT, "f2":DOUBLE], and the operator outputs
 * a column "label" with type STRING, and we want to preserve the column "id", then we get the result
 * schema of ["id":INT, "label":STRING].
 *
 * <p>end user should not directly interact with this helper class. instead it will be indirectly used via concrete algorithms.
 */

• TableUtil
  Utility to operator to interact with Table contents, such as rows and columns.

• VectorTypes 内置Vector类型
  Built-in vector types.

MLEnvironment

The MLEnvironment stores the necessary context in Flink.
Each MLEnvironment will be associated with a unique ID.
The operations associated with the same MLEnvironment ID will share the same Flink job context.

MLEnvironmentFactory

Factory to get the MLEnvironment using a MLEnvironmentId.

operator

• TableSourceBatchOp
  Transform the Table to SourceBatchOp.
• BatchOperator
  Base class of batch algorithm operators.
• TableSourcesStreamOp
  Transform the Table to SourceStreamOp.
• StreamOperator
  Base class of stream algorithm operators.
• AlgoOperatior
  Base class for algorithm operators.

params

• HasOutputCol

• An interface for classes with a parameter specifying the name of the output column.

• HasOutputColDefaultAsNull

• An interface for classes with a parameter specifying name of the output column with a null default value.

• HasOutputCols

• An interface for classes with a parameter specifying names of multiple output columns.

• HasOutputColsDefaultAsNull

• An interface for classes with a parameter specifying names of multiple output columns. The default parameter value is null.

• HasPredictionCol

• An interface for classes with a parameter specifying the column name of the prediction.

• HasPredictionDetailCol

• An interface for classes with a parameter specifying the column name of prediction detail.

• HasReservedCols

• An interface for classes with a parameter specifying the names of the columns to be retained in the output table.

• HasSelectedCol

• An interface for classes with a parameter specifying the name of the table column.

• HasSelectedColDefaultAsNull

• An interface for classes with a parameter specifying the name of the table column with null default value.

• HasSelectedCols

• An interface for classes with a parameter specifying the name of multiple table columns.

• HasSelectedColsDefaultAsNull

• An interface for classes with a parameter specifying the name of multiple table columns with null default value.

• HasMLEnvironmentId

• An interface for classes with a parameter specifying the id of MLEnvironment.

pipeline

• EstimatorBase
  The base class for estimator implementations.

• ModelBase
  The base class for a machine learning model.

• PipelineStageBase
  The base class for a stage in a pipeline, either an [[EstimatorBase]] or a [[TransformerBase]].

• TransformerBase
  The base class for transformer implementations.

下一篇具体分析源码及用法。