spark1.1.0 Machine Learning Library (MLlib)

淩亂°似流年 2022-08-12 01:23 97阅读 0赞

# Machine Learning Library (MLlib) #

MLlib is Spark’s scalable machine learning library consisting of common learning algorithms and utilities, including classification, regression, clustering, collaborative filtering, dimensionality reduction, as well as underlying optimization primitives, as outlined below:

*  [Data types][]
 *  [Basic statistics][]
    
     *  summary statistics
     *  correlations
     *  stratified sampling
     *  hypothesis testing
     *  random data generation
 *  [Classification and regression][]
    
     *  [linear models (SVMs, logistic regression, linear regression)][linear models _SVMs_ logistic regression_ linear regression]
     *  [decision trees][]
     *  [naive Bayes][]
 *  [Collaborative filtering][]
    
     *  alternating least squares (ALS)
 *  [Clustering][]
    
     *  k-means
 *  [Dimensionality reduction][]
    
     *  singular value decomposition (SVD)
     *  principal component analysis (PCA)
 *  [Feature extraction and transformation][]
 *  [Optimization (developer)][Optimization _developer]
    
     *  stochastic gradient descent
     *  limited-memory BFGS (L-BFGS)

MLlib is under active development. The APIs marked `Experimental`/`DeveloperApi` may change in future releases, and the migration guide below will explain all changes between releases.

# Dependencies #

MLlib uses the linear algebra package [Breeze][], which depends on [netlib-java][], and [jblas][]. `netlib-java` and `jblas` depend on native Fortran routines. You need to install the [gfortran runtime library][] if it is not already present on your nodes. MLlib will throw a linking error if it cannot detect these libraries automatically. Due to license issues, we do not include `netlib-java`’s native libraries in MLlib’s dependency set under default settings. If no native library is available at runtime, you will see a warning message. To use native libraries from `netlib-java`, please build Spark with `-Pnetlib-lgpl` or include `com.github.fommil.netlib:all:1.1.2` as a dependency of your project. If you want to use optimized BLAS/LAPACK libraries such as [OpenBLAS][], please link its shared libraries to `/usr/lib/libblas.so.3` and `/usr/lib/liblapack.so.3`, respectively. BLAS/LAPACK libraries on worker nodes should be built without multithreading.

To use MLlib in Python, you will need [NumPy][] version 1.4 or newer.

--------------------

# Migration Guide #

## From 1.0 to 1.1 ##

The only API changes in MLlib v1.1 are in [`DecisionTree`][DecisionTree], which continues to be an experimental API in MLlib 1.1:

1.  *(Breaking change)* The meaning of tree depth has been changed by 1 in order to match the implementations of trees in [scikit-learn][] and in[rpart][]. In MLlib v1.0, a depth-1 tree had 1 leaf node, and a depth-2 tree had 1 root node and 2 leaf nodes. In MLlib v1.1, a depth-0 tree has 1 leaf node, and a depth-1 tree has 1 root node and 2 leaf nodes. This depth is specified by the `maxDepth` parameter in [`Strategy`][Strategy] or via[`DecisionTree`][DecisionTree] static `trainClassifier` and `trainRegressor` methods.
2.  *(Non-breaking change)* We recommend using the newly added `trainClassifier` and `trainRegressor` methods to build a [`DecisionTree`][DecisionTree], rather than using the old parameter class `Strategy`. These new training methods explicitly separate classification and regression, and they replace specialized parameter types with simple `String` types.

Examples of the new, recommended `trainClassifier` and `trainRegressor` are given in the [Decision Trees Guide][].

## From 0.9 to 1.0 ##

In MLlib v1.0, we support both dense and sparse input in a unified way, which introduces a few breaking changes. If your data is sparse, please store it in a sparse format instead of dense to take advantage of sparsity in both storage and computation. Details are described below.

*  [**Scala**][Scala]
 *  [**Java**][Java]
 *  [**Python**][Python]

We used to represent a feature vector by `Array[Double]`, which is replaced by [`Vector`][Vector] in v1.0. Algorithms that used to accept `RDD[Array[Double]]`now take `RDD[Vector]`. [`LabeledPoint`][LabeledPoint] is now a wrapper of `(Double, Vector)` instead of `(Double, Array[Double])`. Converting `Array[Double]` to`Vector` is straightforward:

import org.apache.spark.mllib.linalg.{ Vector, Vectors}
    
    val array: Array[Double] = ... // a double array
    val vector: Vector = Vectors.dense(array) // a dense vector

[`Vectors`][Vectors] provides factory methods to create sparse vectors.

*Note*: Scala imports `scala.collection.immutable.Vector` by default, so you have to import `org.apache.spark.mllib.linalg.Vector` explicitly to use MLlib’s `Vector`.

[Data types]: http://spark.apache.org/docs/latest/mllib-data-types.html
[Basic statistics]: http://spark.apache.org/docs/latest/mllib-statistics.html
[Classification and regression]: http://spark.apache.org/docs/latest/mllib-classification-regression.html
[linear models _SVMs_ logistic regression_ linear regression]: http://spark.apache.org/docs/latest/mllib-linear-methods.html
[decision trees]: http://spark.apache.org/docs/latest/mllib-decision-tree.html
[naive Bayes]: http://spark.apache.org/docs/latest/mllib-naive-bayes.html
[Collaborative filtering]: http://spark.apache.org/docs/latest/mllib-collaborative-filtering.html
[Clustering]: http://spark.apache.org/docs/latest/mllib-clustering.html
[Dimensionality reduction]: http://spark.apache.org/docs/latest/mllib-dimensionality-reduction.html
[Feature extraction and transformation]: http://spark.apache.org/docs/latest/mllib-feature-extraction.html
[Optimization _developer]: http://spark.apache.org/docs/latest/mllib-optimization.html
[Breeze]: http://www.scalanlp.org/
[netlib-java]: https://github.com/fommil/netlib-java
[jblas]: https://github.com/mikiobraun/jblas
[gfortran runtime library]: https://github.com/mikiobraun/jblas/wiki/Missing-Libraries
[OpenBLAS]: http://www.openblas.net/
[NumPy]: http://www.numpy.org/
[DecisionTree]: http://spark.apache.org/docs/latest/api/scala/index.html#org.apache.spark.mllib.tree.DecisionTree
[scikit-learn]: http://scikit-learn.org/stable/modules/classes.html#module-sklearn.tree
[rpart]: http://cran.r-project.org/web/packages/rpart/index.html
[Strategy]: http://spark.apache.org/docs/latest/api/scala/index.html#org.apache.spark.mllib.tree.configuration.Strategy
[Decision Trees Guide]: http://spark.apache.org/docs/latest/mllib-decision-tree.html#examples
[Scala]: http://spark.apache.org/docs/latest/mllib-guide.html#tab_scala_0
[Java]: http://spark.apache.org/docs/latest/mllib-guide.html#tab_java_0
[Python]: http://spark.apache.org/docs/latest/mllib-guide.html#tab_python_0
[Vector]: http://spark.apache.org/docs/latest/api/scala/index.html#org.apache.spark.mllib.linalg.Vector
[LabeledPoint]: http://spark.apache.org/docs/latest/api/scala/index.html#org.apache.spark.mllib.regression.LabeledPoint
[Vectors]: http://spark.apache.org/docs/latest/api/scala/index.html#org.apache.spark.mllib.linalg.Vectors$