-
Notifications
You must be signed in to change notification settings - Fork 16
/
Copy pathRidge-Regression-With-R.html
329 lines (291 loc) · 18.8 KB
/
Ridge-Regression-With-R.html
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
<!DOCTYPE html>
<html>
<head>
<title>Ridge Regression</title>
<meta charset="utf-8">
<meta name="Description" content="R Language Tutorials for Advanced Statistics">
<meta name="Keywords" content="R, Tutorial, Machine learning, Statistics, Data Mining, Analytics, Data science, Linear Regression, Logistic Regression, Time series, Forecasting">
<meta name="Distribution" content="Global">
<meta name="Author" content="Selva Prabhakaran">
<meta name="Robots" content="index, follow">
<meta name="viewport" content="width=device-width, initial-scale=1.0">
<link rel="shortcut icon" href="/screenshots/iconb-64.png" type="image/x-icon" />
<link href="www/bootstrap.min.css" rel="stylesheet">
<link href="www/highlight.css" rel="stylesheet">
<link href='http://fonts.googleapis.com/css?family=Inconsolata:400,700'
rel='stylesheet' type='text/css'>
<!-- Color Script -->
<style type="text/css">
a {
color: #3675C5;
color: rgb(25, 145, 248);
color: #4582ec;
color: #3F73D8;
}
li {
line-height: 1.65;
}
/* reduce spacing around math formula*/
.MathJax_Display {
margin: 0em 0em;
}
</style>
<!-- Add Google search -->
<script language="Javascript" type="text/javascript">
function my_search_google()
{
var query = document.getElementById("my-google-search").value;
window.open("http://google.com/search?q=" + query
+ "%20site:" + "http://r-statistics.co");
}
</script>
</head>
<body>
<div class="container">
<div class="masthead">
<!--
<ul class="nav nav-pills pull-right">
<li class="dropdown">
<a href="#" class="dropdown-toggle" data-toggle="dropdown">
Table of contents<b class="caret"></b>
</a>
<ul class="dropdown-menu pull-right" role="menu">
<li class="dropdown-header"></li>
<li class="dropdown-header">Tutorial</li>
<li><a href="R-Tutorial.html">R Tutorial</a></li>
<li class="dropdown-header">ggplot2</li>
<li><a href="ggplot2-Tutorial-With-R.html">ggplot2 Short Tutorial</a></li>
<li><a href="Complete-Ggplot2-Tutorial-Part1-With-R-Code.html">ggplot2 Tutorial 1 - Intro</a></li>
<li><a href="Complete-Ggplot2-Tutorial-Part2-Customizing-Theme-With-R-Code.html">ggplot2 Tutorial 2 - Theme</a></li>
<li><a href="Top50-Ggplot2-Visualizations-MasterList-R-Code.html">ggplot2 Tutorial 3 - Masterlist</a></li>
<li><a href="ggplot2-cheatsheet.html">ggplot2 Quickref</a></li>
<li class="dropdown-header">Foundations</li>
<li><a href="Linear-Regression.html">Linear Regression</a></li>
<li><a href="Statistical-Tests-in-R.html">Statistical Tests</a></li>
<li><a href="Missing-Value-Treatment-With-R.html">Missing Value Treatment</a></li>
<li><a href="Outlier-Treatment-With-R.html">Outlier Analysis</a></li>
<li><a href="Variable-Selection-and-Importance-With-R.html">Feature Selection</a></li>
<li><a href="Model-Selection-in-R.html">Model Selection</a></li>
<li><a href="Logistic-Regression-With-R.html">Logistic Regression</a></li>
<li><a href="Environments.html">Advanced Linear Regression</a></li>
<li class="dropdown-header">Advanced Regression Models</li>
<li><a href="adv-regression-models.html">Advanced Regression Models</a></li>
<li class="dropdown-header">Time Series</li>
<li><a href="Time-Series-Analysis-With-R.html">Time Series Analysis</a></li>
<li><a href="Time-Series-Forecasting-With-R.html">Time Series Forecasting </a></li>
<li><a href="Time-Series-Forecasting-With-R-part2.html">More Time Series Forecasting</a></li>
<li class="dropdown-header">High Performance Computing</li>
<li><a href="Parallel-Computing-With-R.html">Parallel computing</a></li>
<li><a href="Strategies-To-Improve-And-Speedup-R-Code.html">Strategies to Speedup R code</a></li>
<li class="dropdown-header">Useful Techniques</li>
<li><a href="Association-Mining-With-R.html">Association Mining</a></li>
<li><a href="Multi-Dimensional-Scaling-With-R.html">Multi Dimensional Scaling</a></li>
<li><a href="Profiling.html">Optimization</a></li>
<li><a href="Information-Value-With-R.html">InformationValue package</a></li>
</ul>
</li>
</ul>
-->
<ul class="nav nav-pills pull-right">
<div class="input-group">
<form onsubmit="my_search_google()">
<input type="text" class="form-control" id="my-google-search" placeholder="Search..">
<form>
</div><!-- /input-group -->
</ul><!-- /.col-lg-6 -->
<h3 class="muted"><a href="/">r-statistics.co</a><small> by Selva Prabhakaran</small></h3>
<hr>
</div>
<div class="row">
<div class="col-xs-12 col-sm-3" id="nav">
<div class="well">
<li>
<ul class="list-unstyled">
<li class="dropdown-header"></li>
<li class="dropdown-header">Tutorial</li>
<li><a href="R-Tutorial.html">R Tutorial</a></li>
<li class="dropdown-header">ggplot2</li>
<li><a href="ggplot2-Tutorial-With-R.html">ggplot2 Short Tutorial</a></li>
<li><a href="Complete-Ggplot2-Tutorial-Part1-With-R-Code.html">ggplot2 Tutorial 1 - Intro</a></li>
<li><a href="Complete-Ggplot2-Tutorial-Part2-Customizing-Theme-With-R-Code.html">ggplot2 Tutorial 2 - Theme</a></li>
<li><a href="Top50-Ggplot2-Visualizations-MasterList-R-Code.html">ggplot2 Tutorial 3 - Masterlist</a></li>
<li><a href="ggplot2-cheatsheet.html">ggplot2 Quickref</a></li>
<li class="dropdown-header">Foundations</li>
<li><a href="Linear-Regression.html">Linear Regression</a></li>
<li><a href="Statistical-Tests-in-R.html">Statistical Tests</a></li>
<li><a href="Missing-Value-Treatment-With-R.html">Missing Value Treatment</a></li>
<li><a href="Outlier-Treatment-With-R.html">Outlier Analysis</a></li>
<li><a href="Variable-Selection-and-Importance-With-R.html">Feature Selection</a></li>
<li><a href="Model-Selection-in-R.html">Model Selection</a></li>
<li><a href="Logistic-Regression-With-R.html">Logistic Regression</a></li>
<li><a href="Environments.html">Advanced Linear Regression</a></li>
<li class="dropdown-header">Advanced Regression Models</li>
<li><a href="adv-regression-models.html">Advanced Regression Models</a></li>
<li class="dropdown-header">Time Series</li>
<li><a href="Time-Series-Analysis-With-R.html">Time Series Analysis</a></li>
<li><a href="Time-Series-Forecasting-With-R.html">Time Series Forecasting </a></li>
<li><a href="Time-Series-Forecasting-With-R-part2.html">More Time Series Forecasting</a></li>
<li class="dropdown-header">High Performance Computing</li>
<li><a href="Parallel-Computing-With-R.html">Parallel computing</a></li>
<li><a href="Strategies-To-Improve-And-Speedup-R-Code.html">Strategies to Speedup R code</a></li>
<li class="dropdown-header">Useful Techniques</li>
<li><a href="Association-Mining-With-R.html">Association Mining</a></li>
<li><a href="Multi-Dimensional-Scaling-With-R.html">Multi Dimensional Scaling</a></li>
<li><a href="Profiling.html">Optimization</a></li>
<li><a href="Information-Value-With-R.html">InformationValue package</a></li>
</ul>
</li>
</div>
<div class="well">
<p>Stay up-to-date. <a href="https://docs.google.com/forms/d/1xkMYkLNFU9U39Dd8S_2JC0p8B5t6_Yq6zUQjanQQJpY/viewform">Subscribe!</a></p>
<p><a href="https://docs.google.com/forms/d/13GrkCFcNa-TOIllQghsz2SIEbc-YqY9eJX02B19l5Ow/viewform">Chat!</a></p>
</div>
<h4>Contents</h4>
<ul class="list-unstyled" id="toc"></ul>
<!--
<hr>
<p><a href="/contribute.html">How to contribute</a></p>
<p><a class="btn btn-primary" href="">Edit this page</a></p>
-->
</div>
<div id="content" class="col-xs-12 col-sm-8 pull-right">
<h1>Ridge Regression</h1>
<blockquote>
<p>Ridge Regression is a commonly used technique to address the problem of multi-collinearity. The effectiveness of the application is however debatable.</p>
</blockquote>
<h2>Introduction</h2>
<p>Let us see a use case of the application of Ridge regression on the <code>longley</code> dataset. We will try to predict the <code>GNP.deflator</code> using <code>lm()</code> with the rest of the variables as predictors. This model and results will be compared with the model created using ridge regression.</p>
<div class="sourceCode"><pre class="sourceCode r"><code class="sourceCode r"><span class="kw">library</span> (car) <span class="co"># for VIF</span>
<span class="kw">library</span> (ridge)
<span class="kw">data</span>(longley, <span class="dt">package=</span><span class="st">"datasets"</span>) <span class="co"># initialize data</span>
<span class="kw">head</span> (longley, <span class="dv">4</span>) <span class="co"># show top 4 rows of data</span>
<span class="co">#> GNP.deflator GNP Unemployed Armed.Forces Population Year Employed</span>
<span class="co">#> 1947 83.0 234.289 235.6 159.0 107.608 1947 60.323</span>
<span class="co">#> 1948 88.5 259.426 232.5 145.6 108.632 1948 61.122</span>
<span class="co">#> 1949 88.2 258.054 368.2 161.6 109.773 1949 60.171</span>
<span class="co">#> 1950 89.5 284.599 335.1 165.0 110.929 1950 61.187</span></code></pre></div>
<div class="sourceCode"><pre class="sourceCode r"><code class="sourceCode r">inputData <-<span class="st"> </span><span class="kw">data.frame</span> (longley) <span class="co"># plug in your data here</span>
<span class="kw">colnames</span>(inputData)[<span class="dv">1</span>] <-<span class="st"> "response"</span> <span class="co"># rename response var</span></code></pre></div>
<h2>Calculate Correlations</h2>
<div class="sourceCode"><pre class="sourceCode r"><code class="sourceCode r">XVars <-<span class="st"> </span>inputData[, -<span class="dv">1</span>] <span class="co"># X variables</span>
<span class="kw">round</span>(<span class="kw">cor</span>(XVars), <span class="dv">2</span>) <span class="co"># Correlation Test</span>
<span class="co">#> GNP Unemployed Armed.Forces Population Year Employed</span>
<span class="co">#> GNP 1.00 0.60 0.45 0.99 1.00 0.98</span>
<span class="co">#> Unemployed 0.60 1.00 -0.18 0.69 0.67 0.50</span>
<span class="co">#> Armed.Forces 0.45 -0.18 1.00 0.36 0.42 0.46</span>
<span class="co">#> Population 0.99 0.69 0.36 1.00 0.99 0.96</span>
<span class="co">#> Year 1.00 0.67 0.42 0.99 1.00 0.97</span>
<span class="co">#> Employed 0.98 0.50 0.46 0.96 0.97 1.00</span></code></pre></div>
<h2>Prepare Training And Test Data</h2>
<div class="sourceCode"><pre class="sourceCode r"><code class="sourceCode r"><span class="kw">set.seed</span>(<span class="dv">100</span>) <span class="co"># set seed to replicate results</span>
trainingIndex <-<span class="st"> </span><span class="kw">sample</span>(<span class="dv">1</span>:<span class="kw">nrow</span>(inputData), <span class="fl">0.8</span>*<span class="kw">nrow</span>(inputData)) <span class="co"># indices for 80% training data</span>
trainingData <-<span class="st"> </span>inputData[trainingIndex, ] <span class="co"># training data</span>
testData <-<span class="st"> </span>inputData[-trainingIndex, ] <span class="co"># test data</span></code></pre></div>
<h2>Predict Using Linear Regression</h2>
<div class="sourceCode"><pre class="sourceCode r"><code class="sourceCode r">lmMod <-<span class="st"> </span><span class="kw">lm</span>(response ~<span class="st"> </span>., trainingData) <span class="co"># the linear reg model</span>
<span class="kw">summary</span> (lmMod) <span class="co"># get summary</span>
<span class="kw">vif</span>(lmMod) <span class="co"># get VIF</span>
<span class="co">#> VIF</span>
<span class="co">#> GNP Unemployed Armed.Forces Population Year Employed </span>
<span class="co">#> 1523.74714 93.07635 10.74587 350.58472 2175.29221 182.93609</span>
<span class="co">#> Coefficients:</span>
<span class="co">#> (Intercept) GNP Unemployed Armed.Forces Population Year Employed</span>
<span class="co">#> 7652.25192 0.39214 0.06462 0.01573 -2.33550 -3.83113 0.53060</span></code></pre></div>
<p>There is significant multi-collinearity between GNP & Year and Population & Employed, with negative coefficients in ‘population’ and ‘Employed’. These variables may not contribute much to explain the dependent variable, nevertheless, lets see what this model predicts.</p>
<div class="sourceCode"><pre class="sourceCode r"><code class="sourceCode r">predicted <-<span class="st"> </span><span class="kw">predict</span> (lmMod, testData) <span class="co"># predict on test data</span>
compare <-<span class="st"> </span><span class="kw">cbind</span> (<span class="dt">actual=</span>testData$response, predicted) <span class="co"># combine actual and predicted</span>
<span class="co">#> actual predicted</span>
<span class="co">#> 1949 88.2 88.45501</span>
<span class="co">#> 1953 99.0 96.67492</span>
<span class="co">#> 1957 108.4 106.59672</span>
<span class="co">#> 1959 112.6 113.31106</span>
<span class="kw">mean</span> (<span class="kw">apply</span>(compare, <span class="dv">1</span>, min)/<span class="kw">apply</span>(compare, <span class="dv">1</span>, max)) <span class="co"># calculate accuracy</span>
<span class="co">#> 98.76%</span></code></pre></div>
<h2>Apply Ridge Regression On Same Data</h2>
<div class="sourceCode"><pre class="sourceCode r"><code class="sourceCode r">linRidgeMod <-<span class="st"> </span><span class="kw">linearRidge</span>(response ~<span class="st"> </span>., <span class="dt">data =</span> trainingData) <span class="co"># the ridge regression model</span>
<span class="co">#> No more Negative Coefficients!</span>
<span class="co">#> (Intercept) GNP Unemployed Armed.Forces Population Year Employed</span>
<span class="co">#> -1.015385e+03 3.715498e-02 1.328002e-02 1.707769e-02 1.294903e-01 5.318930e-01 5.976266e-01</span>
predicted <-<span class="st"> </span><span class="kw">predict</span>(linRidgeMod, testData) <span class="co"># predict on test data</span>
compare <-<span class="st"> </span><span class="kw">cbind</span> (<span class="dt">actual=</span>testData$response, predicted) <span class="co"># combine</span>
<span class="co">#> actual predicted</span>
<span class="co">#> 1949 88.2 88.68584</span>
<span class="co">#> 1953 99.0 99.26104</span>
<span class="co">#> 1957 108.4 106.99370</span>
<span class="co">#> 1959 112.6 110.95450</span>
<span class="kw">mean</span> (<span class="kw">apply</span>(compare, <span class="dv">1</span>, min)/<span class="kw">apply</span>(compare, <span class="dv">1</span>, max)) <span class="co"># calculate accuracy</span>
<span class="co">#> 99.10%</span></code></pre></div>
<p>Clearly, in this case, ridge regression is successful in improving the accuracy by a minor but significant fraction.</p>
<h2>Predicting With A Re-calibrated Linear Model</h2>
<div class="sourceCode"><pre class="sourceCode r"><code class="sourceCode r">newlmMod <-<span class="st"> </span><span class="kw">lm</span>(response ~<span class="st"> </span>., trainingData[, -<span class="kw">c</span>(<span class="dv">2</span>, <span class="dv">5</span>, <span class="dv">6</span>)]) <span class="co"># without "GNP", "Population" & "Year"</span>
<span class="kw">summary</span> (newlmMod) <span class="co"># get summary</span>
<span class="kw">vif</span>(newlmMod) <span class="co"># get VIF</span>
<span class="co">#> Coefficients:</span>
<span class="co">#> (Intercept) Unemployed Armed.Forces Employed </span>
<span class="co">#> -62.19771 0.03248 0.02714 2.24039 </span>
<span class="co">#> VIF</span>
<span class="co">#> Unemployed Armed.Forces Employed </span>
<span class="co">#> 2.124153 1.452648 2.592474</span>
predicted <-<span class="st"> </span><span class="kw">predict</span>(newlmMod, testData) <span class="co"># predict on test data</span>
compare <-<span class="st"> </span><span class="kw">cbind</span> (<span class="dt">actual=</span>testData$response, predicted) <span class="co"># for comparison</span>
<span class="kw">mean</span> (<span class="kw">apply</span>(compare, <span class="dv">1</span>, min)/<span class="kw">apply</span>(compare, <span class="dv">1</span>, max)) <span class="co"># calculate accuracy</span>
<span class="co">#> 99.21%</span></code></pre></div>
<p>The re-calibrated linear model yields better accuracy when the multicollinearity is taken care of. This analysis may not be sufficient to draw conclusions about the effectiveness of ridge regression. The intention, however, is to open up considerations for new modeling options for problem solving.</p>
</div>
</div>
<div class="footer">
<hr>
<p>© 2016-17 Selva Prabhakaran. Powered by <a href="http://jekyllrb.com/">jekyll</a>,
<a href="http://yihui.name/knitr/">knitr</a>, and
<a href="http://johnmacfarlane.net/pandoc/">pandoc</a>.
This work is licensed under the <a href="http://creativecommons.org/licenses/by-nc/3.0/">Creative Commons License.</a>
</p>
</div>
</div> <!-- /container -->
<script src="//code.jquery.com/jquery.js"></script>
<script src="www/bootstrap.min.js"></script>
<script src="www/toc.js"></script>
<!-- MathJax Script -->
<script type="text/x-mathjax-config">
MathJax.Hub.Config({
tex2jax: {inlineMath: [['$','$'], ['\\(','\\)']]}
});
</script>
<script type="text/javascript"
src="https://cdn.mathjax.org/mathjax/latest/MathJax.js?config=TeX-AMS-MML_HTMLorMML">
</script>
<!-- Google Analytics Code -->
<script>
(function(i,s,o,g,r,a,m){i['GoogleAnalyticsObject']=r;i[r]=i[r]||function(){
(i[r].q=i[r].q||[]).push(arguments)},i[r].l=1*new Date();a=s.createElement(o),
m=s.getElementsByTagName(o)[0];a.async=1;a.src=g;m.parentNode.insertBefore(a,m)
})(window,document,'script','//www.google-analytics.com/analytics.js','ga');
ga('create', 'UA-69351797-1', 'auto');
ga('send', 'pageview');
</script>
<style type="text/css">
/* reduce spacing around math formula*/
.MathJax_Display {
margin: 0em 0em;
}
body {
font-family: 'Helvetica Neue', Roboto, Arial, sans-serif;
font-size: 16px;
line-height: 27px;
font-weight: 400;
}
blockquote p {
line-height: 1.75;
color: #717171;
}
.well li{
line-height: 28px;
}
li.dropdown-header {
display: block;
padding: 0px;
font-size: 14px;
}
</style>
</body>
</html>