<!DOCTYPE html>
<!--[if lt IE 7]> <html class="no-js lt-ie9 lt-ie8 lt-ie7"> <![endif]-->
<!--[if IE 7]> <html class="no-js lt-ie9 lt-ie8"> <![endif]-->
<!--[if IE 8]> <html class="no-js lt-ie9"> <![endif]-->
<!--[if gt IE 8]><!--> <html class="no-js"> <!--<![endif]-->
<head>
<meta charset="utf-8">
<meta http-equiv="X-UA-Compatible" content="IE=edge,chrome=1">
<title>Optimization - MLlib - Spark 1.5.2 Documentation</title>
<link rel="stylesheet" href="css/bootstrap.min.css">
<style>
body {
padding-top: 60px;
padding-bottom: 40px;
}
</style>
<meta name="viewport" content="width=device-width">
<link rel="stylesheet" href="css/bootstrap-responsive.min.css">
<link rel="stylesheet" href="css/main.css">
<script src="js/vendor/modernizr-2.6.1-respond-1.1.0.min.js"></script>
<link rel="stylesheet" href="css/pygments-default.css">
<!-- Google analytics script -->
<script type="text/javascript">
var _gaq = _gaq || [];
_gaq.push(['_setAccount', 'UA-32518208-2']);
_gaq.push(['_trackPageview']);
(function() {
var ga = document.createElement('script'); ga.type = 'text/javascript'; ga.async = true;
ga.src = ('https:' == document.location.protocol ? 'https://ssl' : 'http://www') + '.google-analytics.com/ga.js';
var s = document.getElementsByTagName('script')[0]; s.parentNode.insertBefore(ga, s);
})();
</script>
</head>
<body>
<!--[if lt IE 7]>
<p class="chromeframe">You are using an outdated browser. <a href="http://browsehappy.com/">Upgrade your browser today</a> or <a href="http://www.google.com/chromeframe/?redirect=true">install Google Chrome Frame</a> to better experience this site.</p>
<![endif]-->
<!-- This code is taken from http://twitter.github.com/bootstrap/examples/hero.html -->
<div class="navbar navbar-fixed-top" id="topbar">
<div class="navbar-inner">
<div class="container">
<div class="brand"><a href="index.html">
<img src="img/spark-logo-hd.png" style="height:50px;"/></a><span class="version">1.5.2</span>
</div>
<ul class="nav">
<!--TODO(andyk): Add class="active" attribute to li some how.-->
<li><a href="index.html">Overview</a></li>
<li class="dropdown">
<a href="#" class="dropdown-toggle" data-toggle="dropdown">Programming Guides<b class="caret"></b></a>
<ul class="dropdown-menu">
<li><a href="quick-start.html">Quick Start</a></li>
<li><a href="programming-guide.html">Spark Programming Guide</a></li>
<li class="divider"></li>
<li><a href="streaming-programming-guide.html">Spark Streaming</a></li>
<li><a href="sql-programming-guide.html">DataFrames and SQL</a></li>
<li><a href="mllib-guide.html">MLlib (Machine Learning)</a></li>
<li><a href="graphx-programming-guide.html">GraphX (Graph Processing)</a></li>
<li><a href="bagel-programming-guide.html">Bagel (Pregel on Spark)</a></li>
<li><a href="sparkr.html">SparkR (R on Spark)</a></li>
</ul>
</li>
<li class="dropdown">
<a href="#" class="dropdown-toggle" data-toggle="dropdown">API Docs<b class="caret"></b></a>
<ul class="dropdown-menu">
<li><a href="api/scala/index.html#org.apache.spark.package">Scala</a></li>
<li><a href="api/java/index.html">Java</a></li>
<li><a href="api/python/index.html">Python</a></li>
<li><a href="api/R/index.html">R</a></li>
</ul>
</li>
<li class="dropdown">
<a href="#" class="dropdown-toggle" data-toggle="dropdown">Deploying<b class="caret"></b></a>
<ul class="dropdown-menu">
<li><a href="cluster-overview.html">Overview</a></li>
<li><a href="submitting-applications.html">Submitting Applications</a></li>
<li class="divider"></li>
<li><a href="spark-standalone.html">Spark Standalone</a></li>
<li><a href="running-on-mesos.html">Mesos</a></li>
<li><a href="running-on-yarn.html">YARN</a></li>
<li class="divider"></li>
<li><a href="ec2-scripts.html">Amazon EC2</a></li>
</ul>
</li>
<li class="dropdown">
<a href="api.html" class="dropdown-toggle" data-toggle="dropdown">More<b class="caret"></b></a>
<ul class="dropdown-menu">
<li><a href="configuration.html">Configuration</a></li>
<li><a href="monitoring.html">Monitoring</a></li>
<li><a href="tuning.html">Tuning Guide</a></li>
<li><a href="job-scheduling.html">Job Scheduling</a></li>
<li><a href="security.html">Security</a></li>
<li><a href="hardware-provisioning.html">Hardware Provisioning</a></li>
<li><a href="hadoop-third-party-distributions.html">3<sup>rd</sup>-Party Hadoop Distros</a></li>
<li class="divider"></li>
<li><a href="building-spark.html">Building Spark</a></li>
<li><a href="https://cwiki.apache.org/confluence/display/SPARK/Contributing+to+Spark">Contributing to Spark</a></li>
<li><a href="https://cwiki.apache.org/confluence/display/SPARK/Supplemental+Spark+Projects">Supplemental Projects</a></li>
</ul>
</li>
</ul>
<!--<p class="navbar-text pull-right"><span class="version-text">v1.5.2</span></p>-->
</div>
</div>
</div>
<div class="container" id="content">
<h1 class="title"><a href="mllib-guide.html">MLlib</a> - Optimization</h1>
<ul id="markdown-toc">
<li><a href="#mathematical-description" id="markdown-toc-mathematical-description">Mathematical description</a> <ul>
<li><a href="#gradient-descent" id="markdown-toc-gradient-descent">Gradient descent</a></li>
<li><a href="#stochastic-gradient-descent-sgd" id="markdown-toc-stochastic-gradient-descent-sgd">Stochastic gradient descent (SGD)</a></li>
<li><a href="#update-schemes-for-distributed-sgd" id="markdown-toc-update-schemes-for-distributed-sgd">Update schemes for distributed SGD</a></li>
<li><a href="#limited-memory-bfgs-l-bfgs" id="markdown-toc-limited-memory-bfgs-l-bfgs">Limited-memory BFGS (L-BFGS)</a></li>
<li><a href="#choosing-an-optimization-method" id="markdown-toc-choosing-an-optimization-method">Choosing an Optimization Method</a></li>
</ul>
</li>
<li><a href="#implementation-in-mllib" id="markdown-toc-implementation-in-mllib">Implementation in MLlib</a> <ul>
<li><a href="#gradient-descent-and-stochastic-gradient-descent" id="markdown-toc-gradient-descent-and-stochastic-gradient-descent">Gradient descent and stochastic gradient descent</a></li>
<li><a href="#l-bfgs" id="markdown-toc-l-bfgs">L-BFGS</a></li>
</ul>
</li>
<li><a href="#developers-notes" id="markdown-toc-developers-notes">Developer’s notes</a></li>
</ul>
<p><code>\[
\newcommand{\R}{\mathbb{R}}
\newcommand{\E}{\mathbb{E}}
\newcommand{\x}{\mathbf{x}}
\newcommand{\y}{\mathbf{y}}
\newcommand{\wv}{\mathbf{w}}
\newcommand{\av}{\mathbf{\alpha}}
\newcommand{\bv}{\mathbf{b}}
\newcommand{\N}{\mathbb{N}}
\newcommand{\id}{\mathbf{I}}
\newcommand{\ind}{\mathbf{1}}
\newcommand{\0}{\mathbf{0}}
\newcommand{\unit}{\mathbf{e}}
\newcommand{\one}{\mathbf{1}}
\newcommand{\zero}{\mathbf{0}}
\]</code></p>
<h2 id="mathematical-description">Mathematical description</h2>
<h3 id="gradient-descent">Gradient descent</h3>
<p>The simplest method to solve optimization problems of the form <code>$\min_{\wv \in\R^d} \; f(\wv)$</code>
is <a href="http://en.wikipedia.org/wiki/Gradient_descent">gradient descent</a>.
Such first-order optimization methods (including gradient descent and stochastic variants
thereof) are well-suited for large-scale and distributed computation.</p>
<p>Gradient descent methods aim to find a local minimum of a function by iteratively taking steps in
the direction of steepest descent, which is the negative of the derivative (called the
<a href="http://en.wikipedia.org/wiki/Gradient">gradient</a>) of the function at the current point, i.e., at
the current parameter value.
If the objective function <code>$f$</code> is not differentiable at all arguments, but still convex, then a
<em>sub-gradient</em>
is the natural generalization of the gradient, and assumes the role of the step direction.
In any case, computing a gradient or sub-gradient of <code>$f$</code> is expensive — it requires a full
pass through the complete dataset, in order to compute the contributions from all loss terms.</p>
<h3 id="stochastic-gradient-descent-sgd">Stochastic gradient descent (SGD)</h3>
<p>Optimization problems whose objective function <code>$f$</code> is written as a sum are particularly
suitable to be solved using <em>stochastic gradient descent (SGD)</em>.
In our case, for the optimization formulations commonly used in <a href="mllib-classification-regression.html">supervised machine learning</a>,
<code>\begin{equation}
f(\wv) :=
\lambda\, R(\wv) +
\frac1n \sum_{i=1}^n L(\wv;\x_i,y_i)
\label{eq:regPrimal}
\ .
\end{equation}</code>
this is especially natural, because the loss is written as an average of the individual losses
coming from each datapoint.</p>
<p>A stochastic subgradient is a randomized choice of a vector, such that in expectation, we obtain
a true subgradient of the original objective function.
Picking one datapoint <code>$i\in[1..n]$</code> uniformly at random, we obtain a stochastic subgradient of
<code>$\eqref{eq:regPrimal}$</code>, with respect to <code>$\wv$</code> as follows:
<code>\[
f'_{\wv,i} := L'_{\wv,i} + \lambda\, R'_\wv \ ,
\]</code>
where <code>$L'_{\wv,i} \in \R^d$</code> is a subgradient of the part of the loss function determined by the
<code>$i$</code>-th datapoint, that is <code>$L'_{\wv,i} \in \frac{\partial}{\partial \wv} L(\wv;\x_i,y_i)$</code>.
Furthermore, <code>$R'_\wv$</code> is a subgradient of the regularizer <code>$R(\wv)$</code>, i.e. <code>$R'_\wv \in
\frac{\partial}{\partial \wv} R(\wv)$</code>. The term <code>$R'_\wv$</code> does not depend on which random
datapoint is picked.
Clearly, in expectation over the random choice of <code>$i\in[1..n]$</code>, we have that <code>$f'_{\wv,i}$</code> is
a subgradient of the original objective <code>$f$</code>, meaning that <code>$\E\left[f'_{\wv,i}\right] \in
\frac{\partial}{\partial \wv} f(\wv)$</code>.</p>
<p>Running SGD now simply becomes walking in the direction of the negative stochastic subgradient
<code>$f'_{\wv,i}$</code>, that is
<code>\begin{equation}\label{eq:SGDupdate}
\wv^{(t+1)} := \wv^{(t)} - \gamma \; f'_{\wv,i} \ .
\end{equation}</code>
<strong>Step-size.</strong>
The parameter <code>$\gamma$</code> is the step-size, which in the default implementation is chosen
decreasing with the square root of the iteration counter, i.e. <code>$\gamma := \frac{s}{\sqrt{t}}$</code>
in the <code>$t$</code>-th iteration, with the input parameter <code>$s=$ stepSize</code>. Note that selecting the best
step-size for SGD methods can often be delicate in practice and is a topic of active research.</p>
<p><strong>Gradients.</strong>
A table of (sub)gradients of the machine learning methods implemented in MLlib, is available in
the <a href="mllib-classification-regression.html">classification and regression</a> section.</p>
<p><strong>Proximal Updates.</strong>
As an alternative to just use the subgradient <code>$R'(\wv)$</code> of the regularizer in the step
direction, an improved update for some cases can be obtained by using the proximal operator
instead.
For the L1-regularizer, the proximal operator is given by soft thresholding, as implemented in
<a href="api/scala/index.html#org.apache.spark.mllib.optimization.L1Updater">L1Updater</a>.</p>
<h3 id="update-schemes-for-distributed-sgd">Update schemes for distributed SGD</h3>
<p>The SGD implementation in
<a href="api/scala/index.html#org.apache.spark.mllib.optimization.GradientDescent">GradientDescent</a> uses
a simple (distributed) sampling of the data examples.
We recall that the loss part of the optimization problem <code>$\eqref{eq:regPrimal}$</code> is
<code>$\frac1n \sum_{i=1}^n L(\wv;\x_i,y_i)$</code>, and therefore <code>$\frac1n \sum_{i=1}^n L'_{\wv,i}$</code> would
be the true (sub)gradient.
Since this would require access to the full data set, the parameter <code>miniBatchFraction</code> specifies
which fraction of the full data to use instead.
The average of the gradients over this subset, i.e.
<code>\[
\frac1{|S|} \sum_{i\in S} L'_{\wv,i} \ ,
\]</code>
is a stochastic gradient. Here <code>$S$</code> is the sampled subset of size <code>$|S|=$ miniBatchFraction
$\cdot n$</code>.</p>
<p>In each iteration, the sampling over the distributed dataset
(<a href="programming-guide.html#resilient-distributed-datasets-rdds">RDD</a>), as well as the
computation of the sum of the partial results from each worker machine is performed by the
standard spark routines.</p>
<p>If the fraction of points <code>miniBatchFraction</code> is set to 1 (default), then the resulting step in
each iteration is exact (sub)gradient descent. In this case there is no randomness and no
variance in the used step directions.
On the other extreme, if <code>miniBatchFraction</code> is chosen very small, such that only a single point
is sampled, i.e. <code>$|S|=$ miniBatchFraction $\cdot n = 1$</code>, then the algorithm is equivalent to
standard SGD. In that case, the step direction depends from the uniformly random sampling of the
point.</p>
<h3 id="limited-memory-bfgs-l-bfgs">Limited-memory BFGS (L-BFGS)</h3>
<p><a href="http://en.wikipedia.org/wiki/Limited-memory_BFGS">L-BFGS</a> is an optimization
algorithm in the family of quasi-Newton methods to solve the optimization problems of the form
<code>$\min_{\wv \in\R^d} \; f(\wv)$</code>. The L-BFGS method approximates the objective function locally as a
quadratic without evaluating the second partial derivatives of the objective function to construct the
Hessian matrix. The Hessian matrix is approximated by previous gradient evaluations, so there is no
vertical scalability issue (the number of training features) when computing the Hessian matrix
explicitly in Newton’s method. As a result, L-BFGS often achieves rapider convergence compared with
other first-order optimization.</p>
<h3 id="choosing-an-optimization-method">Choosing an Optimization Method</h3>
<p><a href="mllib-linear-methods.html">Linear methods</a> use optimization internally, and some linear methods in MLlib support both SGD and L-BFGS.
Different optimization methods can have different convergence guarantees depending on the properties of the objective function, and we cannot cover the literature here.
In general, when L-BFGS is available, we recommend using it instead of SGD since L-BFGS tends to converge faster (in fewer iterations).</p>
<h2 id="implementation-in-mllib">Implementation in MLlib</h2>
<h3 id="gradient-descent-and-stochastic-gradient-descent">Gradient descent and stochastic gradient descent</h3>
<p>Gradient descent methods including stochastic subgradient descent (SGD) as
included as a low-level primitive in <code>MLlib</code>, upon which various ML algorithms
are developed, see the
<a href="mllib-linear-methods.html">linear methods</a>
section for example.</p>
<p>The SGD class
<a href="api/scala/index.html#org.apache.spark.mllib.optimization.GradientDescent">GradientDescent</a>
sets the following parameters:</p>
<ul>
<li><code>Gradient</code> is a class that computes the stochastic gradient of the function
being optimized, i.e., with respect to a single training example, at the
current parameter value. MLlib includes gradient classes for common loss
functions, e.g., hinge, logistic, least-squares. The gradient class takes as
input a training example, its label, and the current parameter value.</li>
<li><code>Updater</code> is a class that performs the actual gradient descent step, i.e.
updating the weights in each iteration, for a given gradient of the loss part.
The updater is also responsible to perform the update from the regularization
part. MLlib includes updaters for cases without regularization, as well as
L1 and L2 regularizers.</li>
<li><code>stepSize</code> is a scalar value denoting the initial step size for gradient
descent. All updaters in MLlib use a step size at the t-th step equal to
<code>stepSize $/ \sqrt{t}$</code>.</li>
<li><code>numIterations</code> is the number of iterations to run.</li>
<li><code>regParam</code> is the regularization parameter when using L1 or L2 regularization.</li>
<li><code>miniBatchFraction</code> is the fraction of the total data that is sampled in
each iteration, to compute the gradient direction.
<ul>
<li>Sampling still requires a pass over the entire RDD, so decreasing <code>miniBatchFraction</code> may not speed up optimization much. Users will see the greatest speedup when the gradient is expensive to compute, for only the chosen samples are used for computing the gradient.</li>
</ul>
</li>
</ul>
<h3 id="l-bfgs">L-BFGS</h3>
<p>L-BFGS is currently only a low-level optimization primitive in <code>MLlib</code>. If you want to use L-BFGS in various
ML algorithms such as Linear Regression, and Logistic Regression, you have to pass the gradient of objective
function, and updater into optimizer yourself instead of using the training APIs like
<a href="api/scala/index.html#org.apache.spark.mllib.classification.LogisticRegressionWithSGD">LogisticRegressionWithSGD</a>.
See the example below. It will be addressed in the next release.</p>
<p>The L1 regularization by using
<a href="api/scala/index.html#org.apache.spark.mllib.optimization.L1Updater">L1Updater</a> will not work since the
soft-thresholding logic in L1Updater is designed for gradient descent. See the developer’s note.</p>
<p>The L-BFGS method
<a href="api/scala/index.html#org.apache.spark.mllib.optimization.LBFGS">LBFGS.runLBFGS</a>
has the following parameters:</p>
<ul>
<li><code>Gradient</code> is a class that computes the gradient of the objective function
being optimized, i.e., with respect to a single training example, at the
current parameter value. MLlib includes gradient classes for common loss
functions, e.g., hinge, logistic, least-squares. The gradient class takes as
input a training example, its label, and the current parameter value.</li>
<li><code>Updater</code> is a class that computes the gradient and loss of objective function
of the regularization part for L-BFGS. MLlib includes updaters for cases without
regularization, as well as L2 regularizer.</li>
<li><code>numCorrections</code> is the number of corrections used in the L-BFGS update. 10 is
recommended.</li>
<li><code>maxNumIterations</code> is the maximal number of iterations that L-BFGS can be run.</li>
<li><code>regParam</code> is the regularization parameter when using regularization.</li>
<li><code>convergenceTol</code> controls how much relative change is still allowed when L-BFGS
is considered to converge. This must be nonnegative. Lower values are less tolerant and
therefore generally cause more iterations to be run. This value looks at both average
improvement and the norm of gradient inside <a href="https://github.com/scalanlp/breeze/blob/master/math/src/main/scala/breeze/optimize/LBFGS.scala">Breeze LBFGS</a>.</li>
</ul>
<p>The <code>return</code> is a tuple containing two elements. The first element is a column matrix
containing weights for every feature, and the second element is an array containing
the loss computed for every iteration.</p>
<p>Here is an example to train binary logistic regression with L2 regularization using
L-BFGS optimizer.</p>
<div class="codetabs">
<div data-lang="scala">
<div class="highlight"><pre><code class="language-scala" data-lang="scala"><span class="k">import</span> <span class="nn">org.apache.spark.SparkContext</span>
<span class="k">import</span> <span class="nn">org.apache.spark.mllib.evaluation.BinaryClassificationMetrics</span>
<span class="k">import</span> <span class="nn">org.apache.spark.mllib.linalg.Vectors</span>
<span class="k">import</span> <span class="nn">org.apache.spark.mllib.util.MLUtils</span>
<span class="k">import</span> <span class="nn">org.apache.spark.mllib.classification.LogisticRegressionModel</span>
<span class="k">import</span> <span class="nn">org.apache.spark.mllib.optimization.</span><span class="o">{</span><span class="nc">LBFGS</span><span class="o">,</span> <span class="nc">LogisticGradient</span><span class="o">,</span> <span class="nc">SquaredL2Updater</span><span class="o">}</span>
<span class="k">val</span> <span class="n">data</span> <span class="k">=</span> <span class="nc">MLUtils</span><span class="o">.</span><span class="n">loadLibSVMFile</span><span class="o">(</span><span class="n">sc</span><span class="o">,</span> <span class="s">"data/mllib/sample_libsvm_data.txt"</span><span class="o">)</span>
<span class="k">val</span> <span class="n">numFeatures</span> <span class="k">=</span> <span class="n">data</span><span class="o">.</span><span class="n">take</span><span class="o">(</span><span class="mi">1</span><span class="o">)(</span><span class="mi">0</span><span class="o">).</span><span class="n">features</span><span class="o">.</span><span class="n">size</span>
<span class="c1">// Split data into training (60%) and test (40%).</span>
<span class="k">val</span> <span class="n">splits</span> <span class="k">=</span> <span class="n">data</span><span class="o">.</span><span class="n">randomSplit</span><span class="o">(</span><span class="nc">Array</span><span class="o">(</span><span class="mf">0.6</span><span class="o">,</span> <span class="mf">0.4</span><span class="o">),</span> <span class="n">seed</span> <span class="k">=</span> <span class="mi">11L</span><span class="o">)</span>
<span class="c1">// Append 1 into the training data as intercept.</span>
<span class="k">val</span> <span class="n">training</span> <span class="k">=</span> <span class="n">splits</span><span class="o">(</span><span class="mi">0</span><span class="o">).</span><span class="n">map</span><span class="o">(</span><span class="n">x</span> <span class="k">=></span> <span class="o">(</span><span class="n">x</span><span class="o">.</span><span class="n">label</span><span class="o">,</span> <span class="nc">MLUtils</span><span class="o">.</span><span class="n">appendBias</span><span class="o">(</span><span class="n">x</span><span class="o">.</span><span class="n">features</span><span class="o">))).</span><span class="n">cache</span><span class="o">()</span>
<span class="k">val</span> <span class="n">test</span> <span class="k">=</span> <span class="n">splits</span><span class="o">(</span><span class="mi">1</span><span class="o">)</span>
<span class="c1">// Run training algorithm to build the model</span>
<span class="k">val</span> <span class="n">numCorrections</span> <span class="k">=</span> <span class="mi">10</span>
<span class="k">val</span> <span class="n">convergenceTol</span> <span class="k">=</span> <span class="mi">1</span><span class="n">e</span><span class="o">-</span><span class="mi">4</span>
<span class="k">val</span> <span class="n">maxNumIterations</span> <span class="k">=</span> <span class="mi">20</span>
<span class="k">val</span> <span class="n">regParam</span> <span class="k">=</span> <span class="mf">0.1</span>
<span class="k">val</span> <span class="n">initialWeightsWithIntercept</span> <span class="k">=</span> <span class="nc">Vectors</span><span class="o">.</span><span class="n">dense</span><span class="o">(</span><span class="k">new</span> <span class="nc">Array</span><span class="o">[</span><span class="kt">Double</span><span class="o">](</span><span class="n">numFeatures</span> <span class="o">+</span> <span class="mi">1</span><span class="o">))</span>
<span class="k">val</span> <span class="o">(</span><span class="n">weightsWithIntercept</span><span class="o">,</span> <span class="n">loss</span><span class="o">)</span> <span class="k">=</span> <span class="nc">LBFGS</span><span class="o">.</span><span class="n">runLBFGS</span><span class="o">(</span>
<span class="n">training</span><span class="o">,</span>
<span class="k">new</span> <span class="nc">LogisticGradient</span><span class="o">(),</span>
<span class="k">new</span> <span class="nc">SquaredL2Updater</span><span class="o">(),</span>
<span class="n">numCorrections</span><span class="o">,</span>
<span class="n">convergenceTol</span><span class="o">,</span>
<span class="n">maxNumIterations</span><span class="o">,</span>
<span class="n">regParam</span><span class="o">,</span>
<span class="n">initialWeightsWithIntercept</span><span class="o">)</span>
<span class="k">val</span> <span class="n">model</span> <span class="k">=</span> <span class="k">new</span> <span class="nc">LogisticRegressionModel</span><span class="o">(</span>
<span class="nc">Vectors</span><span class="o">.</span><span class="n">dense</span><span class="o">(</span><span class="n">weightsWithIntercept</span><span class="o">.</span><span class="n">toArray</span><span class="o">.</span><span class="n">slice</span><span class="o">(</span><span class="mi">0</span><span class="o">,</span> <span class="n">weightsWithIntercept</span><span class="o">.</span><span class="n">size</span> <span class="o">-</span> <span class="mi">1</span><span class="o">)),</span>
<span class="n">weightsWithIntercept</span><span class="o">(</span><span class="n">weightsWithIntercept</span><span class="o">.</span><span class="n">size</span> <span class="o">-</span> <span class="mi">1</span><span class="o">))</span>
<span class="c1">// Clear the default threshold.</span>
<span class="n">model</span><span class="o">.</span><span class="n">clearThreshold</span><span class="o">()</span>
<span class="c1">// Compute raw scores on the test set.</span>
<span class="k">val</span> <span class="n">scoreAndLabels</span> <span class="k">=</span> <span class="n">test</span><span class="o">.</span><span class="n">map</span> <span class="o">{</span> <span class="n">point</span> <span class="k">=></span>
<span class="k">val</span> <span class="n">score</span> <span class="k">=</span> <span class="n">model</span><span class="o">.</span><span class="n">predict</span><span class="o">(</span><span class="n">point</span><span class="o">.</span><span class="n">features</span><span class="o">)</span>
<span class="o">(</span><span class="n">score</span><span class="o">,</span> <span class="n">point</span><span class="o">.</span><span class="n">label</span><span class="o">)</span>
<span class="o">}</span>
<span class="c1">// Get evaluation metrics.</span>
<span class="k">val</span> <span class="n">metrics</span> <span class="k">=</span> <span class="k">new</span> <span class="nc">BinaryClassificationMetrics</span><span class="o">(</span><span class="n">scoreAndLabels</span><span class="o">)</span>
<span class="k">val</span> <span class="n">auROC</span> <span class="k">=</span> <span class="n">metrics</span><span class="o">.</span><span class="n">areaUnderROC</span><span class="o">()</span>
<span class="n">println</span><span class="o">(</span><span class="s">"Loss of each step in training process"</span><span class="o">)</span>
<span class="n">loss</span><span class="o">.</span><span class="n">foreach</span><span class="o">(</span><span class="n">println</span><span class="o">)</span>
<span class="n">println</span><span class="o">(</span><span class="s">"Area under ROC = "</span> <span class="o">+</span> <span class="n">auROC</span><span class="o">)</span></code></pre></div>
</div>
<div data-lang="java">
<div class="highlight"><pre><code class="language-java" data-lang="java"><span class="kn">import</span> <span class="nn">java.util.Arrays</span><span class="o">;</span>
<span class="kn">import</span> <span class="nn">java.util.Random</span><span class="o">;</span>
<span class="kn">import</span> <span class="nn">scala.Tuple2</span><span class="o">;</span>
<span class="kn">import</span> <span class="nn">org.apache.spark.api.java.*</span><span class="o">;</span>
<span class="kn">import</span> <span class="nn">org.apache.spark.api.java.function.Function</span><span class="o">;</span>
<span class="kn">import</span> <span class="nn">org.apache.spark.mllib.classification.LogisticRegressionModel</span><span class="o">;</span>
<span class="kn">import</span> <span class="nn">org.apache.spark.mllib.evaluation.BinaryClassificationMetrics</span><span class="o">;</span>
<span class="kn">import</span> <span class="nn">org.apache.spark.mllib.linalg.Vector</span><span class="o">;</span>
<span class="kn">import</span> <span class="nn">org.apache.spark.mllib.linalg.Vectors</span><span class="o">;</span>
<span class="kn">import</span> <span class="nn">org.apache.spark.mllib.optimization.*</span><span class="o">;</span>
<span class="kn">import</span> <span class="nn">org.apache.spark.mllib.regression.LabeledPoint</span><span class="o">;</span>
<span class="kn">import</span> <span class="nn">org.apache.spark.mllib.util.MLUtils</span><span class="o">;</span>
<span class="kn">import</span> <span class="nn">org.apache.spark.SparkConf</span><span class="o">;</span>
<span class="kn">import</span> <span class="nn">org.apache.spark.SparkContext</span><span class="o">;</span>
<span class="kd">public</span> <span class="kd">class</span> <span class="nc">LBFGSExample</span> <span class="o">{</span>
<span class="kd">public</span> <span class="kd">static</span> <span class="kt">void</span> <span class="nf">main</span><span class="o">(</span><span class="n">String</span><span class="o">[]</span> <span class="n">args</span><span class="o">)</span> <span class="o">{</span>
<span class="n">SparkConf</span> <span class="n">conf</span> <span class="o">=</span> <span class="k">new</span> <span class="nf">SparkConf</span><span class="o">().</span><span class="na">setAppName</span><span class="o">(</span><span class="s">"L-BFGS Example"</span><span class="o">);</span>
<span class="n">SparkContext</span> <span class="n">sc</span> <span class="o">=</span> <span class="k">new</span> <span class="nf">SparkContext</span><span class="o">(</span><span class="n">conf</span><span class="o">);</span>
<span class="n">String</span> <span class="n">path</span> <span class="o">=</span> <span class="s">"data/mllib/sample_libsvm_data.txt"</span><span class="o">;</span>
<span class="n">JavaRDD</span><span class="o"><</span><span class="n">LabeledPoint</span><span class="o">></span> <span class="n">data</span> <span class="o">=</span> <span class="n">MLUtils</span><span class="o">.</span><span class="na">loadLibSVMFile</span><span class="o">(</span><span class="n">sc</span><span class="o">,</span> <span class="n">path</span><span class="o">).</span><span class="na">toJavaRDD</span><span class="o">();</span>
<span class="kt">int</span> <span class="n">numFeatures</span> <span class="o">=</span> <span class="n">data</span><span class="o">.</span><span class="na">take</span><span class="o">(</span><span class="mi">1</span><span class="o">).</span><span class="na">get</span><span class="o">(</span><span class="mi">0</span><span class="o">).</span><span class="na">features</span><span class="o">().</span><span class="na">size</span><span class="o">();</span>
<span class="c1">// Split initial RDD into two... [60% training data, 40% testing data].</span>
<span class="n">JavaRDD</span><span class="o"><</span><span class="n">LabeledPoint</span><span class="o">></span> <span class="n">trainingInit</span> <span class="o">=</span> <span class="n">data</span><span class="o">.</span><span class="na">sample</span><span class="o">(</span><span class="kc">false</span><span class="o">,</span> <span class="mf">0.6</span><span class="o">,</span> <span class="mi">11L</span><span class="o">);</span>
<span class="n">JavaRDD</span><span class="o"><</span><span class="n">LabeledPoint</span><span class="o">></span> <span class="n">test</span> <span class="o">=</span> <span class="n">data</span><span class="o">.</span><span class="na">subtract</span><span class="o">(</span><span class="n">trainingInit</span><span class="o">);</span>
<span class="c1">// Append 1 into the training data as intercept.</span>
<span class="n">JavaRDD</span><span class="o"><</span><span class="n">Tuple2</span><span class="o"><</span><span class="n">Object</span><span class="o">,</span> <span class="n">Vector</span><span class="o">>></span> <span class="n">training</span> <span class="o">=</span> <span class="n">data</span><span class="o">.</span><span class="na">map</span><span class="o">(</span>
<span class="k">new</span> <span class="n">Function</span><span class="o"><</span><span class="n">LabeledPoint</span><span class="o">,</span> <span class="n">Tuple2</span><span class="o"><</span><span class="n">Object</span><span class="o">,</span> <span class="n">Vector</span><span class="o">>>()</span> <span class="o">{</span>
<span class="kd">public</span> <span class="n">Tuple2</span><span class="o"><</span><span class="n">Object</span><span class="o">,</span> <span class="n">Vector</span><span class="o">></span> <span class="nf">call</span><span class="o">(</span><span class="n">LabeledPoint</span> <span class="n">p</span><span class="o">)</span> <span class="o">{</span>
<span class="k">return</span> <span class="k">new</span> <span class="n">Tuple2</span><span class="o"><</span><span class="n">Object</span><span class="o">,</span> <span class="n">Vector</span><span class="o">>(</span><span class="n">p</span><span class="o">.</span><span class="na">label</span><span class="o">(),</span> <span class="n">MLUtils</span><span class="o">.</span><span class="na">appendBias</span><span class="o">(</span><span class="n">p</span><span class="o">.</span><span class="na">features</span><span class="o">()));</span>
<span class="o">}</span>
<span class="o">});</span>
<span class="n">training</span><span class="o">.</span><span class="na">cache</span><span class="o">();</span>
<span class="c1">// Run training algorithm to build the model.</span>
<span class="kt">int</span> <span class="n">numCorrections</span> <span class="o">=</span> <span class="mi">10</span><span class="o">;</span>
<span class="kt">double</span> <span class="n">convergenceTol</span> <span class="o">=</span> <span class="mi">1</span><span class="n">e</span><span class="o">-</span><span class="mi">4</span><span class="o">;</span>
<span class="kt">int</span> <span class="n">maxNumIterations</span> <span class="o">=</span> <span class="mi">20</span><span class="o">;</span>
<span class="kt">double</span> <span class="n">regParam</span> <span class="o">=</span> <span class="mf">0.1</span><span class="o">;</span>
<span class="n">Vector</span> <span class="n">initialWeightsWithIntercept</span> <span class="o">=</span> <span class="n">Vectors</span><span class="o">.</span><span class="na">dense</span><span class="o">(</span><span class="k">new</span> <span class="kt">double</span><span class="o">[</span><span class="n">numFeatures</span> <span class="o">+</span> <span class="mi">1</span><span class="o">]);</span>
<span class="n">Tuple2</span><span class="o"><</span><span class="n">Vector</span><span class="o">,</span> <span class="kt">double</span><span class="o">[]></span> <span class="n">result</span> <span class="o">=</span> <span class="n">LBFGS</span><span class="o">.</span><span class="na">runLBFGS</span><span class="o">(</span>
<span class="n">training</span><span class="o">.</span><span class="na">rdd</span><span class="o">(),</span>
<span class="k">new</span> <span class="nf">LogisticGradient</span><span class="o">(),</span>
<span class="k">new</span> <span class="nf">SquaredL2Updater</span><span class="o">(),</span>
<span class="n">numCorrections</span><span class="o">,</span>
<span class="n">convergenceTol</span><span class="o">,</span>
<span class="n">maxNumIterations</span><span class="o">,</span>
<span class="n">regParam</span><span class="o">,</span>
<span class="n">initialWeightsWithIntercept</span><span class="o">);</span>
<span class="n">Vector</span> <span class="n">weightsWithIntercept</span> <span class="o">=</span> <span class="n">result</span><span class="o">.</span><span class="na">_1</span><span class="o">();</span>
<span class="kt">double</span><span class="o">[]</span> <span class="n">loss</span> <span class="o">=</span> <span class="n">result</span><span class="o">.</span><span class="na">_2</span><span class="o">();</span>
<span class="kd">final</span> <span class="n">LogisticRegressionModel</span> <span class="n">model</span> <span class="o">=</span> <span class="k">new</span> <span class="nf">LogisticRegressionModel</span><span class="o">(</span>
<span class="n">Vectors</span><span class="o">.</span><span class="na">dense</span><span class="o">(</span><span class="n">Arrays</span><span class="o">.</span><span class="na">copyOf</span><span class="o">(</span><span class="n">weightsWithIntercept</span><span class="o">.</span><span class="na">toArray</span><span class="o">(),</span> <span class="n">weightsWithIntercept</span><span class="o">.</span><span class="na">size</span><span class="o">()</span> <span class="o">-</span> <span class="mi">1</span><span class="o">)),</span>
<span class="o">(</span><span class="n">weightsWithIntercept</span><span class="o">.</span><span class="na">toArray</span><span class="o">())[</span><span class="n">weightsWithIntercept</span><span class="o">.</span><span class="na">size</span><span class="o">()</span> <span class="o">-</span> <span class="mi">1</span><span class="o">]);</span>
<span class="c1">// Clear the default threshold.</span>
<span class="n">model</span><span class="o">.</span><span class="na">clearThreshold</span><span class="o">();</span>
<span class="c1">// Compute raw scores on the test set.</span>
<span class="n">JavaRDD</span><span class="o"><</span><span class="n">Tuple2</span><span class="o"><</span><span class="n">Object</span><span class="o">,</span> <span class="n">Object</span><span class="o">>></span> <span class="n">scoreAndLabels</span> <span class="o">=</span> <span class="n">test</span><span class="o">.</span><span class="na">map</span><span class="o">(</span>
<span class="k">new</span> <span class="n">Function</span><span class="o"><</span><span class="n">LabeledPoint</span><span class="o">,</span> <span class="n">Tuple2</span><span class="o"><</span><span class="n">Object</span><span class="o">,</span> <span class="n">Object</span><span class="o">>>()</span> <span class="o">{</span>
<span class="kd">public</span> <span class="n">Tuple2</span><span class="o"><</span><span class="n">Object</span><span class="o">,</span> <span class="n">Object</span><span class="o">></span> <span class="nf">call</span><span class="o">(</span><span class="n">LabeledPoint</span> <span class="n">p</span><span class="o">)</span> <span class="o">{</span>
<span class="n">Double</span> <span class="n">score</span> <span class="o">=</span> <span class="n">model</span><span class="o">.</span><span class="na">predict</span><span class="o">(</span><span class="n">p</span><span class="o">.</span><span class="na">features</span><span class="o">());</span>
<span class="k">return</span> <span class="k">new</span> <span class="n">Tuple2</span><span class="o"><</span><span class="n">Object</span><span class="o">,</span> <span class="n">Object</span><span class="o">>(</span><span class="n">score</span><span class="o">,</span> <span class="n">p</span><span class="o">.</span><span class="na">label</span><span class="o">());</span>
<span class="o">}</span>
<span class="o">});</span>
<span class="c1">// Get evaluation metrics.</span>
<span class="n">BinaryClassificationMetrics</span> <span class="n">metrics</span> <span class="o">=</span>
<span class="k">new</span> <span class="nf">BinaryClassificationMetrics</span><span class="o">(</span><span class="n">scoreAndLabels</span><span class="o">.</span><span class="na">rdd</span><span class="o">());</span>
<span class="kt">double</span> <span class="n">auROC</span> <span class="o">=</span> <span class="n">metrics</span><span class="o">.</span><span class="na">areaUnderROC</span><span class="o">();</span>
<span class="n">System</span><span class="o">.</span><span class="na">out</span><span class="o">.</span><span class="na">println</span><span class="o">(</span><span class="s">"Loss of each step in training process"</span><span class="o">);</span>
<span class="k">for</span> <span class="o">(</span><span class="kt">double</span> <span class="n">l</span> <span class="o">:</span> <span class="n">loss</span><span class="o">)</span>
<span class="n">System</span><span class="o">.</span><span class="na">out</span><span class="o">.</span><span class="na">println</span><span class="o">(</span><span class="n">l</span><span class="o">);</span>
<span class="n">System</span><span class="o">.</span><span class="na">out</span><span class="o">.</span><span class="na">println</span><span class="o">(</span><span class="s">"Area under ROC = "</span> <span class="o">+</span> <span class="n">auROC</span><span class="o">);</span>
<span class="o">}</span>
<span class="o">}</span></code></pre></div>
</div>
</div>
<h2 id="developers-notes">Developer’s notes</h2>
<p>Since the Hessian is constructed approximately from previous gradient evaluations,
the objective function can not be changed during the optimization process.
As a result, Stochastic L-BFGS will not work naively by just using miniBatch;
therefore, we don’t provide this until we have better understanding.</p>
<p><code>Updater</code> is a class originally designed for gradient decent which computes
the actual gradient descent step. However, we’re able to take the gradient and
loss of objective function of regularization for L-BFGS by ignoring the part of logic
only for gradient decent such as adaptive step size stuff. We will refactorize
this into regularizer to replace updater to separate the logic between
regularization and step update later.</p>
</div> <!-- /container -->
<script src="js/vendor/jquery-1.8.0.min.js"></script>
<script src="js/vendor/bootstrap.min.js"></script>
<script src="js/vendor/anchor.min.js"></script>
<script src="js/main.js"></script>
<!-- MathJax Section -->
<script type="text/x-mathjax-config">
MathJax.Hub.Config({
TeX: { equationNumbers: { autoNumber: "AMS" } }
});
</script>
<script>
// Note that we load MathJax this way to work with local file (file://), HTTP and HTTPS.
// We could use "//cdn.mathjax...", but that won't support "file://".
(function(d, script) {
script = d.createElement('script');
script.type = 'text/javascript';
script.async = true;
script.onload = function(){
MathJax.Hub.Config({
tex2jax: {
inlineMath: [ ["$", "$"], ["\\\\(","\\\\)"] ],
displayMath: [ ["$$","$$"], ["\\[", "\\]"] ],
processEscapes: true,
skipTags: ['script', 'noscript', 'style', 'textarea', 'pre']
}
});
};
script.src = ('https:' == document.location.protocol ? 'https://' : 'http://') +
'cdn.mathjax.org/mathjax/latest/MathJax.js?config=TeX-AMS-MML_HTMLorMML';
d.getElementsByTagName('head')[0].appendChild(script);
}(document));
</script>
</body>
</html>