<!DOCTYPE html PUBLIC "-//W3C//DTD XHTML 1.0 Transitional//EN"
"http://www.w3.org/TR/xhtml1/DTD/xhtml1-transitional.dtd">
<html xmlns="http://www.w3.org/1999/xhtml">
<head>
<meta http-equiv="Content-Type" content="text/html; charset=utf-8" />
<title>1.10. Decision Trees — scikit-learn 0.21.3 documentation</title>
<!-- htmltitle is before nature.css - we use this hack to load bootstrap first -->
<meta name="viewport" content="width=device-width, initial-scale=1.0" />
<link rel="stylesheet" href="../_static/css/bootstrap.min.css" media="screen" />
<link rel="stylesheet" href="../_static/css/bootstrap-responsive.css"/>
<link rel="stylesheet" href="../_static/nature.css" type="text/css" />
<link rel="stylesheet" href="../_static/pygments.css" type="text/css" />
<link rel="stylesheet" href="../_static/gallery.css" type="text/css" />
<script type="text/javascript">
var DOCUMENTATION_OPTIONS = {
URL_ROOT: '../',
VERSION: '0.21.3',
COLLAPSE_INDEX: false,
FILE_SUFFIX: '.html',
HAS_SOURCE: true,
SOURCELINK_SUFFIX: '.txt'
};
</script>
<script type="text/javascript" src="../_static/jquery.js"></script>
<script type="text/javascript" src="../_static/underscore.js"></script>
<script type="text/javascript" src="../_static/doctools.js"></script>
<script type="text/javascript" src="../_static/js/copybutton.js"></script>
<script type="text/javascript" src="../_static/js/extra.js"></script>
<script type="text/javascript" src="https://cdnjs.cloudflare.com/ajax/libs/mathjax/2.7.0/MathJax.js?config=TeX-AMS_SVG"></script>
<link rel="shortcut icon" href="../_static/favicon.ico"/>
<link rel="author" title="About these documents" href="../about.html" />
<link rel="search" title="Search" href="../search.html" />
<link rel="next" title="1.11. Ensemble methods" href="ensemble.html" />
<link rel="prev" title="1.9. Naive Bayes" href="naive_bayes.html" />
<meta name="viewport" content="width=device-width, initial-scale=1.0" />
<script src="../_static/js/bootstrap.min.js" type="text/javascript"></script>
<script>
VERSION_SUBDIR = (function(groups) {
return groups ? groups[1] : null;
})(location.href.match(/^https?:\/\/scikit-learn.org\/([^\/]+)/));
</script>
<link rel="canonical" href="https://scikit-learn.org/stable/modules/tree.html" />
<script type="text/javascript">
$("div.buttonNext, div.buttonPrevious").hover(
function () {
$(this).css('background-color', '#FF9C34');
},
function () {
$(this).css('background-color', '#A7D6E2');
}
);
function showMenu() {
var topNav = document.getElementById("scikit-navbar");
if (topNav.className === "navbar") {
topNav.className += " responsive";
} else {
topNav.className = "navbar";
}
};
</script>
</head>
<body>
<div class="header-wrapper">
<div class="header">
<p class="logo"><a href="../index.html">
<img src="../_static/scikit-learn-logo-small.png" alt="Logo"/>
</a>
</p><div class="navbar" id="scikit-navbar">
<ul>
<li><a href="../index.html">Home</a></li>
<li><a href="../install.html">Installation</a></li>
<li class="btn-li"><div class="btn-group">
<a href="../documentation.html">Documentation</a>
<a class="btn dropdown-toggle" data-toggle="dropdown">
<span class="caret"></span>
</a>
<ul class="dropdown-menu">
<li class="link-title">Scikit-learn <script>document.write(DOCUMENTATION_OPTIONS.VERSION + (VERSION_SUBDIR ? " (" + VERSION_SUBDIR + ")" : ""));</script></li>
<li><a href="../tutorial/index.html">Tutorials</a></li>
<li><a href="../user_guide.html">User guide</a></li>
<li><a href="classes.html">API</a></li>
<li><a href="../glossary.html">Glossary</a></li>
<li><a href="../faq.html">FAQ</a></li>
<li><a href="../developers/index.html">Development</a></li>
<li><a href="../roadmap.html">Roadmap</a></li>
<li><a href="../about.html">About us</a></li>
<li class="divider"></li>
<script>if (VERSION_SUBDIR != "stable") document.write('<li><a href="http://scikit-learn.org/stable/documentation.html">Stable version</a></li>')</script>
<script>if (VERSION_SUBDIR != "dev") document.write('<li><a href="http://scikit-learn.org/dev/documentation.html">Development version</a></li>')</script>
<li><a href="http://scikit-learn.org/dev/versions.html">All available versions</a></li>
<li><a href="../_downloads/scikit-learn-docs.pdf">PDF documentation</a></li>
</ul>
</div>
</li>
<li><a href="../auto_examples/index.html">Examples</a></li>
</ul>
<a href="javascript:void(0);" onclick="showMenu()">
<div class="nav-icon">
<div class="hamburger-line"></div>
<div class="hamburger-line"></div>
<div class="hamburger-line"></div>
</div>
</a>
<div class="search_form">
<div class="gcse-search" id="cse" style="width: 100%;"></div>
</div>
</div> <!-- end navbar --></div>
</div>
<!-- GitHub "fork me" ribbon -->
<a href="https://github.com/scikit-learn/scikit-learn">
<img class="fork-me"
style="position: absolute; top: 0; right: 0; border: 0;"
src="../_static/img/forkme.png"
alt="Fork me on GitHub" />
</a>
<div class="content-wrapper">
<div class="sphinxsidebar">
<div class="sphinxsidebarwrapper">
<div class="rel">
<div class="rellink">
<a href="naive_bayes.html"
accesskey="P">Previous
<br/>
<span class="smallrellink">
1.9. Naive Bayes
</span>
<span class="hiddenrellink">
1.9. Naive Bayes
</span>
</a>
</div>
<div class="spacer">
</div>
<div class="rellink">
<a href="ensemble.html"
accesskey="N">Next
<br/>
<span class="smallrellink">
1.11. Ensembl...
</span>
<span class="hiddenrellink">
1.11. Ensemble methods
</span>
</a>
</div>
<!-- Ad a link to the 'up' page -->
<div class="spacer">
</div>
<div class="rellink">
<a href="../supervised_learning.html">
Up
<br/>
<span class="smallrellink">
1. Supervised...
</span>
<span class="hiddenrellink">
1. Supervised learning
</span>
</a>
</div>
</div>
<p class="doc-version"><b>scikit-learn v0.21.3</b><br/>
<a href="http://scikit-learn.org/dev/versions.html">Other versions</a></p>
<p class="citing">Please <b><a href="../about.html#citing-scikit-learn" style="font-size: 110%;">cite us </a></b>if you use the software.</p>
<ul>
<li><a class="reference internal" href="#">1.10. Decision Trees</a><ul>
<li><a class="reference internal" href="#classification">1.10.1. Classification</a></li>
<li><a class="reference internal" href="#regression">1.10.2. Regression</a></li>
<li><a class="reference internal" href="#multi-output-problems">1.10.3. Multi-output problems</a></li>
<li><a class="reference internal" href="#complexity">1.10.4. Complexity</a></li>
<li><a class="reference internal" href="#tips-on-practical-use">1.10.5. Tips on practical use</a></li>
<li><a class="reference internal" href="#tree-algorithms-id3-c4-5-c5-0-and-cart">1.10.6. Tree algorithms: ID3, C4.5, C5.0 and CART</a></li>
<li><a class="reference internal" href="#mathematical-formulation">1.10.7. Mathematical formulation</a><ul>
<li><a class="reference internal" href="#classification-criteria">1.10.7.1. Classification criteria</a></li>
<li><a class="reference internal" href="#regression-criteria">1.10.7.2. Regression criteria</a></li>
</ul>
</li>
</ul>
</li>
</ul>
</div>
</div>
<input type="checkbox" id="nav-trigger" class="nav-trigger" checked />
<label for="nav-trigger"></label>
<div class="content">
<div class="documentwrapper">
<div class="bodywrapper">
<div class="body" role="main">
<div class="section" id="decision-trees">
<span id="tree"></span><h1>1.10. Decision Trees<a class="headerlink" href="#decision-trees" title="Permalink to this headline">¶</a></h1>
<p><strong>Decision Trees (DTs)</strong> are a non-parametric supervised learning method used
for <a class="reference internal" href="#tree-classification"><span class="std std-ref">classification</span></a> and <a class="reference internal" href="#tree-regression"><span class="std std-ref">regression</span></a>. The goal is to create a model that predicts the value of a
target variable by learning simple decision rules inferred from the data
features.</p>
<p>For instance, in the example below, decision trees learn from data to
approximate a sine curve with a set of if-then-else decision rules. The deeper
the tree, the more complex the decision rules and the fitter the model.</p>
<div class="figure align-center">
<a class="reference external image-reference" href="../auto_examples/tree/plot_tree_regression.html"><img alt="../_images/sphx_glr_plot_tree_regression_0011.png" src="../_images/sphx_glr_plot_tree_regression_0011.png" style="width: 480.0px; height: 360.0px;" /></a>
</div>
<p>Some advantages of decision trees are:</p>
<blockquote>
<div><ul class="simple">
<li>Simple to understand and to interpret. Trees can be visualised.</li>
<li>Requires little data preparation. Other techniques often require data
normalisation, dummy variables need to be created and blank values to
be removed. Note however that this module does not support missing
values.</li>
<li>The cost of using the tree (i.e., predicting data) is logarithmic in the
number of data points used to train the tree.</li>
<li>Able to handle both numerical and categorical data. Other techniques
are usually specialised in analysing datasets that have only one type
of variable. See <a class="reference internal" href="#tree-algorithms"><span class="std std-ref">algorithms</span></a> for more
information.</li>
<li>Able to handle multi-output problems.</li>
<li>Uses a white box model. If a given situation is observable in a model,
the explanation for the condition is easily explained by boolean logic.
By contrast, in a black box model (e.g., in an artificial neural
network), results may be more difficult to interpret.</li>
<li>Possible to validate a model using statistical tests. That makes it
possible to account for the reliability of the model.</li>
<li>Performs well even if its assumptions are somewhat violated by
the true model from which the data were generated.</li>
</ul>
</div></blockquote>
<p>The disadvantages of decision trees include:</p>
<blockquote>
<div><ul class="simple">
<li>Decision-tree learners can create over-complex trees that do not
generalise the data well. This is called overfitting. Mechanisms
such as pruning (not currently supported), setting the minimum
number of samples required at a leaf node or setting the maximum
depth of the tree are necessary to avoid this problem.</li>
<li>Decision trees can be unstable because small variations in the
data might result in a completely different tree being generated.
This problem is mitigated by using decision trees within an
ensemble.</li>
<li>The problem of learning an optimal decision tree is known to be
NP-complete under several aspects of optimality and even for simple
concepts. Consequently, practical decision-tree learning algorithms
are based on heuristic algorithms such as the greedy algorithm where
locally optimal decisions are made at each node. Such algorithms
cannot guarantee to return the globally optimal decision tree. This
can be mitigated by training multiple trees in an ensemble learner,
where the features and samples are randomly sampled with replacement.</li>
<li>There are concepts that are hard to learn because decision trees
do not express them easily, such as XOR, parity or multiplexer problems.</li>
<li>Decision tree learners create biased trees if some classes dominate.
It is therefore recommended to balance the dataset prior to fitting
with the decision tree.</li>
</ul>
</div></blockquote>
<div class="section" id="classification">
<span id="tree-classification"></span><h2>1.10.1. Classification<a class="headerlink" href="#classification" title="Permalink to this headline">¶</a></h2>
<p><a class="reference internal" href="generated/sklearn.tree.DecisionTreeClassifier.html#sklearn.tree.DecisionTreeClassifier" title="sklearn.tree.DecisionTreeClassifier"><code class="xref py py-class docutils literal"><span class="pre">DecisionTreeClassifier</span></code></a> is a class capable of performing multi-class
classification on a dataset.</p>
<p>As with other classifiers, <a class="reference internal" href="generated/sklearn.tree.DecisionTreeClassifier.html#sklearn.tree.DecisionTreeClassifier" title="sklearn.tree.DecisionTreeClassifier"><code class="xref py py-class docutils literal"><span class="pre">DecisionTreeClassifier</span></code></a> takes as input two arrays:
an array X, sparse or dense, of size <code class="docutils literal"><span class="pre">[n_samples,</span> <span class="pre">n_features]</span></code> holding the
training samples, and an array Y of integer values, size <code class="docutils literal"><span class="pre">[n_samples]</span></code>,
holding the class labels for the training samples:</p>
<div class="highlight-default"><div class="highlight"><pre><span></span><span class="gp">>>> </span><span class="kn">from</span> <span class="nn">sklearn</span> <span class="k">import</span> <span class="n">tree</span>
<span class="gp">>>> </span><span class="n">X</span> <span class="o">=</span> <span class="p">[[</span><span class="mi">0</span><span class="p">,</span> <span class="mi">0</span><span class="p">],</span> <span class="p">[</span><span class="mi">1</span><span class="p">,</span> <span class="mi">1</span><span class="p">]]</span>
<span class="gp">>>> </span><span class="n">Y</span> <span class="o">=</span> <span class="p">[</span><span class="mi">0</span><span class="p">,</span> <span class="mi">1</span><span class="p">]</span>
<span class="gp">>>> </span><span class="n">clf</span> <span class="o">=</span> <span class="n">tree</span><span class="o">.</span><span class="n">DecisionTreeClassifier</span><span class="p">()</span>
<span class="gp">>>> </span><span class="n">clf</span> <span class="o">=</span> <span class="n">clf</span><span class="o">.</span><span class="n">fit</span><span class="p">(</span><span class="n">X</span><span class="p">,</span> <span class="n">Y</span><span class="p">)</span>
</pre></div>
</div>
<p>After being fitted, the model can then be used to predict the class of samples:</p>
<div class="highlight-default"><div class="highlight"><pre><span></span><span class="gp">>>> </span><span class="n">clf</span><span class="o">.</span><span class="n">predict</span><span class="p">([[</span><span class="mf">2.</span><span class="p">,</span> <span class="mf">2.</span><span class="p">]])</span>
<span class="go">array([1])</span>
</pre></div>
</div>
<p>Alternatively, the probability of each class can be predicted, which is the
fraction of training samples of the same class in a leaf:</p>
<div class="highlight-default"><div class="highlight"><pre><span></span><span class="gp">>>> </span><span class="n">clf</span><span class="o">.</span><span class="n">predict_proba</span><span class="p">([[</span><span class="mf">2.</span><span class="p">,</span> <span class="mf">2.</span><span class="p">]])</span>
<span class="go">array([[0., 1.]])</span>
</pre></div>
</div>
<p><a class="reference internal" href="generated/sklearn.tree.DecisionTreeClassifier.html#sklearn.tree.DecisionTreeClassifier" title="sklearn.tree.DecisionTreeClassifier"><code class="xref py py-class docutils literal"><span class="pre">DecisionTreeClassifier</span></code></a> is capable of both binary (where the
labels are [-1, 1]) classification and multiclass (where the labels are
[0, …, K-1]) classification.</p>
<p>Using the Iris dataset, we can construct a tree as follows:</p>
<div class="highlight-default"><div class="highlight"><pre><span></span><span class="gp">>>> </span><span class="kn">from</span> <span class="nn">sklearn.datasets</span> <span class="k">import</span> <span class="n">load_iris</span>
<span class="gp">>>> </span><span class="kn">from</span> <span class="nn">sklearn</span> <span class="k">import</span> <span class="n">tree</span>
<span class="gp">>>> </span><span class="n">iris</span> <span class="o">=</span> <span class="n">load_iris</span><span class="p">()</span>
<span class="gp">>>> </span><span class="n">clf</span> <span class="o">=</span> <span class="n">tree</span><span class="o">.</span><span class="n">DecisionTreeClassifier</span><span class="p">()</span>
<span class="gp">>>> </span><span class="n">clf</span> <span class="o">=</span> <span class="n">clf</span><span class="o">.</span><span class="n">fit</span><span class="p">(</span><span class="n">iris</span><span class="o">.</span><span class="n">data</span><span class="p">,</span> <span class="n">iris</span><span class="o">.</span><span class="n">target</span><span class="p">)</span>
</pre></div>
</div>
<p>Once trained, you can plot the tree with the plot_tree function:</p>
<div class="highlight-default"><div class="highlight"><pre><span></span><span class="gp">>>> </span><span class="n">tree</span><span class="o">.</span><span class="n">plot_tree</span><span class="p">(</span><span class="n">clf</span><span class="o">.</span><span class="n">fit</span><span class="p">(</span><span class="n">iris</span><span class="o">.</span><span class="n">data</span><span class="p">,</span> <span class="n">iris</span><span class="o">.</span><span class="n">target</span><span class="p">))</span>
</pre></div>
</div>
<div class="figure align-center">
<a class="reference external image-reference" href="../auto_examples/tree/plot_iris_dtc.html"><img alt="../_images/sphx_glr_plot_iris_dtc_0021.png" src="../_images/sphx_glr_plot_iris_dtc_0021.png" style="width: 480.0px; height: 360.0px;" /></a>
</div>
<p>We can also export the tree in <a class="reference external" href="https://www.graphviz.org/">Graphviz</a> format using the <a class="reference internal" href="generated/sklearn.tree.export_graphviz.html#sklearn.tree.export_graphviz" title="sklearn.tree.export_graphviz"><code class="xref py py-func docutils literal"><span class="pre">export_graphviz</span></code></a>
exporter. If you use the <a class="reference external" href="https://conda.io">conda</a> package manager, the graphviz binaries</p>
<p>and the python package can be installed with</p>
<blockquote>
<div>conda install python-graphviz</div></blockquote>
<p>Alternatively binaries for graphviz can be downloaded from the graphviz project homepage,
and the Python wrapper installed from pypi with <code class="docutils literal"><span class="pre">pip</span> <span class="pre">install</span> <span class="pre">graphviz</span></code>.</p>
<p>Below is an example graphviz export of the above tree trained on the entire
iris dataset; the results are saved in an output file <code class="docutils literal"><span class="pre">iris.pdf</span></code>:</p>
<div class="highlight-default"><div class="highlight"><pre><span></span><span class="gp">>>> </span><span class="kn">import</span> <span class="nn">graphviz</span>
<span class="gp">>>> </span><span class="n">dot_data</span> <span class="o">=</span> <span class="n">tree</span><span class="o">.</span><span class="n">export_graphviz</span><span class="p">(</span><span class="n">clf</span><span class="p">,</span> <span class="n">out_file</span><span class="o">=</span><span class="kc">None</span><span class="p">)</span>
<span class="gp">>>> </span><span class="n">graph</span> <span class="o">=</span> <span class="n">graphviz</span><span class="o">.</span><span class="n">Source</span><span class="p">(</span><span class="n">dot_data</span><span class="p">)</span>
<span class="gp">>>> </span><span class="n">graph</span><span class="o">.</span><span class="n">render</span><span class="p">(</span><span class="s2">"iris"</span><span class="p">)</span>
</pre></div>
</div>
<p>The <a class="reference internal" href="generated/sklearn.tree.export_graphviz.html#sklearn.tree.export_graphviz" title="sklearn.tree.export_graphviz"><code class="xref py py-func docutils literal"><span class="pre">export_graphviz</span></code></a> exporter also supports a variety of aesthetic
options, including coloring nodes by their class (or value for regression) and
using explicit variable and class names if desired. Jupyter notebooks also
render these plots inline automatically:</p>
<div class="highlight-default"><div class="highlight"><pre><span></span><span class="gp">>>> </span><span class="n">dot_data</span> <span class="o">=</span> <span class="n">tree</span><span class="o">.</span><span class="n">export_graphviz</span><span class="p">(</span><span class="n">clf</span><span class="p">,</span> <span class="n">out_file</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span>
<span class="gp">... </span> <span class="n">feature_names</span><span class="o">=</span><span class="n">iris</span><span class="o">.</span><span class="n">feature_names</span><span class="p">,</span>
<span class="gp">... </span> <span class="n">class_names</span><span class="o">=</span><span class="n">iris</span><span class="o">.</span><span class="n">target_names</span><span class="p">,</span>
<span class="gp">... </span> <span class="n">filled</span><span class="o">=</span><span class="kc">True</span><span class="p">,</span> <span class="n">rounded</span><span class="o">=</span><span class="kc">True</span><span class="p">,</span>
<span class="gp">... </span> <span class="n">special_characters</span><span class="o">=</span><span class="kc">True</span><span class="p">)</span>
<span class="gp">>>> </span><span class="n">graph</span> <span class="o">=</span> <span class="n">graphviz</span><span class="o">.</span><span class="n">Source</span><span class="p">(</span><span class="n">dot_data</span><span class="p">)</span>
<span class="gp">>>> </span><span class="n">graph</span>
</pre></div>
</div>
<div class="figure align-center">
<img alt="../_images/iris.png" src="../_images/iris.png" />
</div>
<div class="figure align-center">
<a class="reference external image-reference" href="../auto_examples/tree/plot_iris_dtc.html"><img alt="../_images/sphx_glr_plot_iris_dtc_0011.png" src="../_images/sphx_glr_plot_iris_dtc_0011.png" style="width: 480.0px; height: 360.0px;" /></a>
</div>
<p>Alternatively, the tree can also be exported in textual format with the
function <a class="reference internal" href="generated/sklearn.tree.export_text.html#sklearn.tree.export_text" title="sklearn.tree.export_text"><code class="xref py py-func docutils literal"><span class="pre">export_text</span></code></a>. This method doesn’t require the installation
of external libraries and is more compact:</p>
<div class="highlight-default"><div class="highlight"><pre><span></span><span class="gp">>>> </span><span class="kn">from</span> <span class="nn">sklearn.datasets</span> <span class="k">import</span> <span class="n">load_iris</span>
<span class="gp">>>> </span><span class="kn">from</span> <span class="nn">sklearn.tree</span> <span class="k">import</span> <span class="n">DecisionTreeClassifier</span>
<span class="gp">>>> </span><span class="kn">from</span> <span class="nn">sklearn.tree.export</span> <span class="k">import</span> <span class="n">export_text</span>
<span class="gp">>>> </span><span class="n">iris</span> <span class="o">=</span> <span class="n">load_iris</span><span class="p">()</span>
<span class="gp">>>> </span><span class="n">X</span> <span class="o">=</span> <span class="n">iris</span><span class="p">[</span><span class="s1">'data'</span><span class="p">]</span>
<span class="gp">>>> </span><span class="n">y</span> <span class="o">=</span> <span class="n">iris</span><span class="p">[</span><span class="s1">'target'</span><span class="p">]</span>
<span class="gp">>>> </span><span class="n">decision_tree</span> <span class="o">=</span> <span class="n">DecisionTreeClassifier</span><span class="p">(</span><span class="n">random_state</span><span class="o">=</span><span class="mi">0</span><span class="p">,</span> <span class="n">max_depth</span><span class="o">=</span><span class="mi">2</span><span class="p">)</span>
<span class="gp">>>> </span><span class="n">decision_tree</span> <span class="o">=</span> <span class="n">decision_tree</span><span class="o">.</span><span class="n">fit</span><span class="p">(</span><span class="n">X</span><span class="p">,</span> <span class="n">y</span><span class="p">)</span>
<span class="gp">>>> </span><span class="n">r</span> <span class="o">=</span> <span class="n">export_text</span><span class="p">(</span><span class="n">decision_tree</span><span class="p">,</span> <span class="n">feature_names</span><span class="o">=</span><span class="n">iris</span><span class="p">[</span><span class="s1">'feature_names'</span><span class="p">])</span>
<span class="gp">>>> </span><span class="nb">print</span><span class="p">(</span><span class="n">r</span><span class="p">)</span>
<span class="go">|--- petal width (cm) <= 0.80</span>
<span class="go">| |--- class: 0</span>
<span class="go">|--- petal width (cm) > 0.80</span>
<span class="go">| |--- petal width (cm) <= 1.75</span>
<span class="go">| | |--- class: 1</span>
<span class="go">| |--- petal width (cm) > 1.75</span>
<span class="go">| | |--- class: 2</span>
</pre></div>
</div>
<div class="topic">
<p class="topic-title first">Examples:</p>
<ul class="simple">
<li><a class="reference internal" href="../auto_examples/tree/plot_iris_dtc.html#sphx-glr-auto-examples-tree-plot-iris-dtc-py"><span class="std std-ref">Plot the decision surface of a decision tree on the iris dataset</span></a></li>
<li><a class="reference internal" href="../auto_examples/tree/plot_unveil_tree_structure.html#sphx-glr-auto-examples-tree-plot-unveil-tree-structure-py"><span class="std std-ref">Understanding the decision tree structure</span></a></li>
</ul>
</div>
</div>
<div class="section" id="regression">
<span id="tree-regression"></span><h2>1.10.2. Regression<a class="headerlink" href="#regression" title="Permalink to this headline">¶</a></h2>
<div class="figure align-center">
<a class="reference external image-reference" href="../auto_examples/tree/plot_tree_regression.html"><img alt="../_images/sphx_glr_plot_tree_regression_0011.png" src="../_images/sphx_glr_plot_tree_regression_0011.png" style="width: 480.0px; height: 360.0px;" /></a>
</div>
<p>Decision trees can also be applied to regression problems, using the
<a class="reference internal" href="generated/sklearn.tree.DecisionTreeRegressor.html#sklearn.tree.DecisionTreeRegressor" title="sklearn.tree.DecisionTreeRegressor"><code class="xref py py-class docutils literal"><span class="pre">DecisionTreeRegressor</span></code></a> class.</p>
<p>As in the classification setting, the fit method will take as argument arrays X
and y, only that in this case y is expected to have floating point values
instead of integer values:</p>
<div class="highlight-default"><div class="highlight"><pre><span></span><span class="gp">>>> </span><span class="kn">from</span> <span class="nn">sklearn</span> <span class="k">import</span> <span class="n">tree</span>
<span class="gp">>>> </span><span class="n">X</span> <span class="o">=</span> <span class="p">[[</span><span class="mi">0</span><span class="p">,</span> <span class="mi">0</span><span class="p">],</span> <span class="p">[</span><span class="mi">2</span><span class="p">,</span> <span class="mi">2</span><span class="p">]]</span>
<span class="gp">>>> </span><span class="n">y</span> <span class="o">=</span> <span class="p">[</span><span class="mf">0.5</span><span class="p">,</span> <span class="mf">2.5</span><span class="p">]</span>
<span class="gp">>>> </span><span class="n">clf</span> <span class="o">=</span> <span class="n">tree</span><span class="o">.</span><span class="n">DecisionTreeRegressor</span><span class="p">()</span>
<span class="gp">>>> </span><span class="n">clf</span> <span class="o">=</span> <span class="n">clf</span><span class="o">.</span><span class="n">fit</span><span class="p">(</span><span class="n">X</span><span class="p">,</span> <span class="n">y</span><span class="p">)</span>
<span class="gp">>>> </span><span class="n">clf</span><span class="o">.</span><span class="n">predict</span><span class="p">([[</span><span class="mi">1</span><span class="p">,</span> <span class="mi">1</span><span class="p">]])</span>
<span class="go">array([0.5])</span>
</pre></div>
</div>
<div class="topic">
<p class="topic-title first">Examples:</p>
<ul class="simple">
<li><a class="reference internal" href="../auto_examples/tree/plot_tree_regression.html#sphx-glr-auto-examples-tree-plot-tree-regression-py"><span class="std std-ref">Decision Tree Regression</span></a></li>
</ul>
</div>
</div>
<div class="section" id="multi-output-problems">
<span id="tree-multioutput"></span><h2>1.10.3. Multi-output problems<a class="headerlink" href="#multi-output-problems" title="Permalink to this headline">¶</a></h2>
<p>A multi-output problem is a supervised learning problem with several outputs
to predict, that is when Y is a 2d array of size <code class="docutils literal"><span class="pre">[n_samples,</span> <span class="pre">n_outputs]</span></code>.</p>
<p>When there is no correlation between the outputs, a very simple way to solve
this kind of problem is to build n independent models, i.e. one for each
output, and then to use those models to independently predict each one of the n
outputs. However, because it is likely that the output values related to the
same input are themselves correlated, an often better way is to build a single
model capable of predicting simultaneously all n outputs. First, it requires
lower training time since only a single estimator is built. Second, the
generalization accuracy of the resulting estimator may often be increased.</p>
<p>With regard to decision trees, this strategy can readily be used to support
multi-output problems. This requires the following changes:</p>
<blockquote>
<div><ul class="simple">
<li>Store n output values in leaves, instead of 1;</li>
<li>Use splitting criteria that compute the average reduction across all
n outputs.</li>
</ul>
</div></blockquote>
<p>This module offers support for multi-output problems by implementing this
strategy in both <a class="reference internal" href="generated/sklearn.tree.DecisionTreeClassifier.html#sklearn.tree.DecisionTreeClassifier" title="sklearn.tree.DecisionTreeClassifier"><code class="xref py py-class docutils literal"><span class="pre">DecisionTreeClassifier</span></code></a> and
<a class="reference internal" href="generated/sklearn.tree.DecisionTreeRegressor.html#sklearn.tree.DecisionTreeRegressor" title="sklearn.tree.DecisionTreeRegressor"><code class="xref py py-class docutils literal"><span class="pre">DecisionTreeRegressor</span></code></a>. If a decision tree is fit on an output array Y
of size <code class="docutils literal"><span class="pre">[n_samples,</span> <span class="pre">n_outputs]</span></code> then the resulting estimator will:</p>
<blockquote>
<div><ul class="simple">
<li>Output n_output values upon <code class="docutils literal"><span class="pre">predict</span></code>;</li>
<li>Output a list of n_output arrays of class probabilities upon
<code class="docutils literal"><span class="pre">predict_proba</span></code>.</li>
</ul>
</div></blockquote>
<p>The use of multi-output trees for regression is demonstrated in
<a class="reference internal" href="../auto_examples/tree/plot_tree_regression_multioutput.html#sphx-glr-auto-examples-tree-plot-tree-regression-multioutput-py"><span class="std std-ref">Multi-output Decision Tree Regression</span></a>. In this example, the input
X is a single real value and the outputs Y are the sine and cosine of X.</p>
<div class="figure align-center">
<a class="reference external image-reference" href="../auto_examples/tree/plot_tree_regression_multioutput.html"><img alt="../_images/sphx_glr_plot_tree_regression_multioutput_0011.png" src="../_images/sphx_glr_plot_tree_regression_multioutput_0011.png" style="width: 480.0px; height: 360.0px;" /></a>
</div>
<p>The use of multi-output trees for classification is demonstrated in
<a class="reference internal" href="../auto_examples/plot_multioutput_face_completion.html#sphx-glr-auto-examples-plot-multioutput-face-completion-py"><span class="std std-ref">Face completion with a multi-output estimators</span></a>. In this example, the inputs
X are the pixels of the upper half of faces and the outputs Y are the pixels of
the lower half of those faces.</p>
<div class="figure align-center">
<a class="reference external image-reference" href="../auto_examples/plot_multioutput_face_completion.html"><img alt="../_images/sphx_glr_plot_multioutput_face_completion_0011.png" src="../_images/sphx_glr_plot_multioutput_face_completion_0011.png" style="width: 750.0px; height: 847.5px;" /></a>
</div>
<div class="topic">
<p class="topic-title first">Examples:</p>
<ul class="simple">
<li><a class="reference internal" href="../auto_examples/tree/plot_tree_regression_multioutput.html#sphx-glr-auto-examples-tree-plot-tree-regression-multioutput-py"><span class="std std-ref">Multi-output Decision Tree Regression</span></a></li>
<li><a class="reference internal" href="../auto_examples/plot_multioutput_face_completion.html#sphx-glr-auto-examples-plot-multioutput-face-completion-py"><span class="std std-ref">Face completion with a multi-output estimators</span></a></li>
</ul>
</div>
<div class="topic">
<p class="topic-title first">References:</p>
<ul class="simple">
<li>M. Dumont et al, <a class="reference external" href="http://www.montefiore.ulg.ac.be/services/stochastic/pubs/2009/DMWG09/dumont-visapp09-shortpaper.pdf">Fast multi-class image annotation with random subwindows
and multiple output randomized trees</a>, International Conference on
Computer Vision Theory and Applications 2009</li>
</ul>
</div>
</div>
<div class="section" id="complexity">
<span id="tree-complexity"></span><h2>1.10.4. Complexity<a class="headerlink" href="#complexity" title="Permalink to this headline">¶</a></h2>
<p>In general, the run time cost to construct a balanced binary tree is
<span class="math">\(O(n_{samples}n_{features}\log(n_{samples}))\)</span> and query time
<span class="math">\(O(\log(n_{samples}))\)</span>. Although the tree construction algorithm attempts
to generate balanced trees, they will not always be balanced. Assuming that the
subtrees remain approximately balanced, the cost at each node consists of
searching through <span class="math">\(O(n_{features})\)</span> to find the feature that offers the
largest reduction in entropy. This has a cost of
<span class="math">\(O(n_{features}n_{samples}\log(n_{samples}))\)</span> at each node, leading to a
total cost over the entire trees (by summing the cost at each node) of
<span class="math">\(O(n_{features}n_{samples}^{2}\log(n_{samples}))\)</span>.</p>
<p>Scikit-learn offers a more efficient implementation for the construction of
decision trees. A naive implementation (as above) would recompute the class
label histograms (for classification) or the means (for regression) at for each
new split point along a given feature. Presorting the feature over all
relevant samples, and retaining a running label count, will reduce the complexity
at each node to <span class="math">\(O(n_{features}\log(n_{samples}))\)</span>, which results in a
total cost of <span class="math">\(O(n_{features}n_{samples}\log(n_{samples}))\)</span>. This is an option
for all tree based algorithms. By default it is turned on for gradient boosting,
where in general it makes training faster, but turned off for all other algorithms as
it tends to slow down training when training deep trees.</p>
</div>
<div class="section" id="tips-on-practical-use">
<h2>1.10.5. Tips on practical use<a class="headerlink" href="#tips-on-practical-use" title="Permalink to this headline">¶</a></h2>
<blockquote>
<div><ul class="simple">
<li>Decision trees tend to overfit on data with a large number of features.
Getting the right ratio of samples to number of features is important, since
a tree with few samples in high dimensional space is very likely to overfit.</li>
<li>Consider performing dimensionality reduction (<a class="reference internal" href="decomposition.html#pca"><span class="std std-ref">PCA</span></a>,
<a class="reference internal" href="decomposition.html#ica"><span class="std std-ref">ICA</span></a>, or <a class="reference internal" href="feature_selection.html#feature-selection"><span class="std std-ref">Feature selection</span></a>) beforehand to
give your tree a better chance of finding features that are discriminative.</li>
<li><a class="reference internal" href="../auto_examples/tree/plot_unveil_tree_structure.html#sphx-glr-auto-examples-tree-plot-unveil-tree-structure-py"><span class="std std-ref">Understanding the decision tree structure</span></a> will help
in gaining more insights about how the decision tree makes predictions, which is
important for understanding the important features in the data.</li>
<li>Visualise your tree as you are training by using the <code class="docutils literal"><span class="pre">export</span></code>
function. Use <code class="docutils literal"><span class="pre">max_depth=3</span></code> as an initial tree depth to get a feel for
how the tree is fitting to your data, and then increase the depth.</li>
<li>Remember that the number of samples required to populate the tree doubles
for each additional level the tree grows to. Use <code class="docutils literal"><span class="pre">max_depth</span></code> to control
the size of the tree to prevent overfitting.</li>
<li>Use <code class="docutils literal"><span class="pre">min_samples_split</span></code> or <code class="docutils literal"><span class="pre">min_samples_leaf</span></code> to ensure that multiple
samples inform every decision in the tree, by controlling which splits will
be considered. A very small number will usually mean the tree will overfit,
whereas a large number will prevent the tree from learning the data. Try
<code class="docutils literal"><span class="pre">min_samples_leaf=5</span></code> as an initial value. If the sample size varies
greatly, a float number can be used as percentage in these two parameters.
While <code class="docutils literal"><span class="pre">min_samples_split</span></code> can create arbitrarily small leaves,
<code class="docutils literal"><span class="pre">min_samples_leaf</span></code> guarantees that each leaf has a minimum size, avoiding
low-variance, over-fit leaf nodes in regression problems. For
classification with few classes, <code class="docutils literal"><span class="pre">min_samples_leaf=1</span></code> is often the best
choice.</li>
<li>Balance your dataset before training to prevent the tree from being biased
toward the classes that are dominant. Class balancing can be done by
sampling an equal number of samples from each class, or preferably by
normalizing the sum of the sample weights (<code class="docutils literal"><span class="pre">sample_weight</span></code>) for each
class to the same value. Also note that weight-based pre-pruning criteria,
such as <code class="docutils literal"><span class="pre">min_weight_fraction_leaf</span></code>, will then be less biased toward
dominant classes than criteria that are not aware of the sample weights,
like <code class="docutils literal"><span class="pre">min_samples_leaf</span></code>.</li>
<li>If the samples are weighted, it will be easier to optimize the tree
structure using weight-based pre-pruning criterion such as
<code class="docutils literal"><span class="pre">min_weight_fraction_leaf</span></code>, which ensure that leaf nodes contain at least
a fraction of the overall sum of the sample weights.</li>
<li>All decision trees use <code class="docutils literal"><span class="pre">np.float32</span></code> arrays internally.
If training data is not in this format, a copy of the dataset will be made.</li>
<li>If the input matrix X is very sparse, it is recommended to convert to sparse
<code class="docutils literal"><span class="pre">csc_matrix</span></code> before calling fit and sparse <code class="docutils literal"><span class="pre">csr_matrix</span></code> before calling
predict. Training time can be orders of magnitude faster for a sparse
matrix input compared to a dense matrix when features have zero values in
most of the samples.</li>
</ul>
</div></blockquote>
</div>
<div class="section" id="tree-algorithms-id3-c4-5-c5-0-and-cart">
<span id="tree-algorithms"></span><h2>1.10.6. Tree algorithms: ID3, C4.5, C5.0 and CART<a class="headerlink" href="#tree-algorithms-id3-c4-5-c5-0-and-cart" title="Permalink to this headline">¶</a></h2>
<p>What are all the various decision tree algorithms and how do they differ
from each other? Which one is implemented in scikit-learn?</p>
<p><a class="reference external" href="https://en.wikipedia.org/wiki/ID3_algorithm">ID3</a> (Iterative Dichotomiser 3) was developed in 1986 by Ross Quinlan.
The algorithm creates a multiway tree, finding for each node (i.e. in
a greedy manner) the categorical feature that will yield the largest
information gain for categorical targets. Trees are grown to their
maximum size and then a pruning step is usually applied to improve the
ability of the tree to generalise to unseen data.</p>
<p>C4.5 is the successor to ID3 and removed the restriction that features
must be categorical by dynamically defining a discrete attribute (based
on numerical variables) that partitions the continuous attribute value
into a discrete set of intervals. C4.5 converts the trained trees
(i.e. the output of the ID3 algorithm) into sets of if-then rules.
These accuracy of each rule is then evaluated to determine the order
in which they should be applied. Pruning is done by removing a rule’s
precondition if the accuracy of the rule improves without it.</p>
<p>C5.0 is Quinlan’s latest version release under a proprietary license.
It uses less memory and builds smaller rulesets than C4.5 while being
more accurate.</p>
<p><a class="reference external" href="https://en.wikipedia.org/wiki/Predictive_analytics#Classification_and_regression_trees_.28CART.29">CART</a> (Classification and Regression Trees) is very similar to C4.5, but
it differs in that it supports numerical target variables (regression) and
does not compute rule sets. CART constructs binary trees using the feature
and threshold that yield the largest information gain at each node.</p>
<p>scikit-learn uses an optimised version of the CART algorithm; however, scikit-learn
implementation does not support categorical variables for now.</p>
</div>
<div class="section" id="mathematical-formulation">
<span id="tree-mathematical-formulation"></span><h2>1.10.7. Mathematical formulation<a class="headerlink" href="#mathematical-formulation" title="Permalink to this headline">¶</a></h2>
<p>Given training vectors <span class="math">\(x_i \in R^n\)</span>, i=1,…, l and a label vector
<span class="math">\(y \in R^l\)</span>, a decision tree recursively partitions the space such
that the samples with the same labels are grouped together.</p>
<p>Let the data at node <span class="math">\(m\)</span> be represented by <span class="math">\(Q\)</span>. For
each candidate split <span class="math">\(\theta = (j, t_m)\)</span> consisting of a
feature <span class="math">\(j\)</span> and threshold <span class="math">\(t_m\)</span>, partition the data into
<span class="math">\(Q_{left}(\theta)\)</span> and <span class="math">\(Q_{right}(\theta)\)</span> subsets</p>
<div class="math">
\[ \begin{align}\begin{aligned}Q_{left}(\theta) = {(x, y) | x_j <= t_m}\\Q_{right}(\theta) = Q \setminus Q_{left}(\theta)\end{aligned}\end{align} \]</div>
<p>The impurity at <span class="math">\(m\)</span> is computed using an impurity function
<span class="math">\(H()\)</span>, the choice of which depends on the task being solved
(classification or regression)</p>
<div class="math">
\[G(Q, \theta) = \frac{n_{left}}{N_m} H(Q_{left}(\theta))
+ \frac{n_{right}}{N_m} H(Q_{right}(\theta))\]</div>
<p>Select the parameters that minimises the impurity</p>
<div class="math">
\[\theta^* = \operatorname{argmin}_\theta G(Q, \theta)\]</div>
<p>Recurse for subsets <span class="math">\(Q_{left}(\theta^*)\)</span> and
<span class="math">\(Q_{right}(\theta^*)\)</span> until the maximum allowable depth is reached,
<span class="math">\(N_m < \min_{samples}\)</span> or <span class="math">\(N_m = 1\)</span>.</p>
<div class="section" id="classification-criteria">
<h3>1.10.7.1. Classification criteria<a class="headerlink" href="#classification-criteria" title="Permalink to this headline">¶</a></h3>
<p>If a target is a classification outcome taking on values 0,1,…,K-1,
for node <span class="math">\(m\)</span>, representing a region <span class="math">\(R_m\)</span> with <span class="math">\(N_m\)</span>
observations, let</p>
<div class="math">
\[p_{mk} = 1/ N_m \sum_{x_i \in R_m} I(y_i = k)\]</div>
<p>be the proportion of class k observations in node <span class="math">\(m\)</span></p>
<p>Common measures of impurity are Gini</p>
<div class="math">
\[H(X_m) = \sum_k p_{mk} (1 - p_{mk})\]</div>
<p>Entropy</p>
<div class="math">
\[H(X_m) = - \sum_k p_{mk} \log(p_{mk})\]</div>
<p>and Misclassification</p>
<div class="math">
\[H(X_m) = 1 - \max(p_{mk})\]</div>
<p>where <span class="math">\(X_m\)</span> is the training data in node <span class="math">\(m\)</span></p>
</div>
<div class="section" id="regression-criteria">
<h3>1.10.7.2. Regression criteria<a class="headerlink" href="#regression-criteria" title="Permalink to this headline">¶</a></h3>
<p>If the target is a continuous value, then for node <span class="math">\(m\)</span>,
representing a region <span class="math">\(R_m\)</span> with <span class="math">\(N_m\)</span> observations, common
criteria to minimise as for determining locations for future
splits are Mean Squared Error, which minimizes the L2 error
using mean values at terminal nodes, and Mean Absolute Error, which
minimizes the L1 error using median values at terminal nodes.</p>
<p>Mean Squared Error:</p>
<div class="math">
\[ \begin{align}\begin{aligned}\bar{y}_m = \frac{1}{N_m} \sum_{i \in N_m} y_i\\H(X_m) = \frac{1}{N_m} \sum_{i \in N_m} (y_i - \bar{y}_m)^2\end{aligned}\end{align} \]</div>
<p>Mean Absolute Error:</p>
<div class="math">
\[ \begin{align}\begin{aligned}\bar{y}_m = \frac{1}{N_m} \sum_{i \in N_m} y_i\\H(X_m) = \frac{1}{N_m} \sum_{i \in N_m} |y_i - \bar{y}_m|\end{aligned}\end{align} \]</div>
<p>where <span class="math">\(X_m\)</span> is the training data in node <span class="math">\(m\)</span></p>
<div class="topic">
<p class="topic-title first">References:</p>
<ul class="simple">
<li><a class="reference external" href="https://en.wikipedia.org/wiki/Decision_tree_learning">https://en.wikipedia.org/wiki/Decision_tree_learning</a></li>
<li><a class="reference external" href="https://en.wikipedia.org/wiki/Predictive_analytics">https://en.wikipedia.org/wiki/Predictive_analytics</a></li>
<li>L. Breiman, J. Friedman, R. Olshen, and C. Stone. Classification and
Regression Trees. Wadsworth, Belmont, CA, 1984.</li>
<li>J.R. Quinlan. C4. 5: programs for machine learning. Morgan Kaufmann, 1993.</li>
<li>T. Hastie, R. Tibshirani and J. Friedman.
Elements of Statistical Learning, Springer, 2009.</li>
</ul>
</div>
</div>
</div>
</div>
</div>
</div>
</div>
<div class="clearer"></div>
</div>
</div>
<div class="footer">
© 2007 - 2019, scikit-learn developers (BSD License).
<a href="../_sources/modules/tree.rst.txt" rel="nofollow">Show this page source</a>
</div>
<div class="rel">
<div class="buttonPrevious">
<a href="naive_bayes.html">Previous
</a>
</div>
<div class="buttonNext">
<a href="ensemble.html">Next
</a>
</div>
</div>
<script>
window.ga=window.ga||function(){(ga.q=ga.q||[]).push(arguments)};ga.l=+new Date;
ga('create', 'UA-22606712-2', 'auto');
ga('set', 'anonymizeIp', true);
ga('send', 'pageview');
</script>
<script async src='https://www.google-analytics.com/analytics.js'></script>
<script>
(function() {
var cx = '016639176250731907682:tjtqbvtvij0';
var gcse = document.createElement('script'); gcse.type = 'text/javascript'; gcse.async = true;
gcse.src = 'https://cse.google.com/cse.js?cx=' + cx;
var s = document.getElementsByTagName('script')[0]; s.parentNode.insertBefore(gcse, s);
})();
</script>
<script src="https://scikit-learn.org/versionwarning.js"></script>
<script src="https://scikit-learn.org/versionwarning.js"></script>
</body>
</html>