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
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
|
<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01 Transitional//EN" "http://www.w3.org/TR/html4/loose.dtd">
<!-- NewPage -->
<html lang="en">
<head>
<!-- Generated by javadoc (version 1.7.0_51) on Wed Sep 16 15:55:12 PDT 2015 -->
<title>LogisticGradient</title>
<meta name="date" content="2015-09-16">
<link rel="stylesheet" type="text/css" href="../../../../../stylesheet.css" title="Style">
</head>
<body>
<script type="text/javascript"><!--
if (location.href.indexOf('is-external=true') == -1) {
parent.document.title="LogisticGradient";
}
//-->
</script>
<noscript>
<div>JavaScript is disabled on your browser.</div>
</noscript>
<!-- ========= START OF TOP NAVBAR ======= -->
<div class="topNav"><a name="navbar_top">
<!-- -->
</a><a href="#skip-navbar_top" title="Skip navigation links"></a><a name="navbar_top_firstrow">
<!-- -->
</a>
<ul class="navList" title="Navigation">
<li><a href="../../../../../overview-summary.html">Overview</a></li>
<li><a href="package-summary.html">Package</a></li>
<li class="navBarCell1Rev">Class</li>
<li><a href="package-tree.html">Tree</a></li>
<li><a href="../../../../../deprecated-list.html">Deprecated</a></li>
<li><a href="../../../../../index-all.html">Index</a></li>
<li><a href="../../../../../help-doc.html">Help</a></li>
</ul>
</div>
<div class="subNav">
<ul class="navList">
<li><a href="../../../../../org/apache/spark/mllib/optimization/LeastSquaresGradient.html" title="class in org.apache.spark.mllib.optimization"><span class="strong">Prev Class</span></a></li>
<li><a href="../../../../../org/apache/spark/mllib/optimization/Optimizer.html" title="interface in org.apache.spark.mllib.optimization"><span class="strong">Next Class</span></a></li>
</ul>
<ul class="navList">
<li><a href="../../../../../index.html?org/apache/spark/mllib/optimization/LogisticGradient.html" target="_top">Frames</a></li>
<li><a href="LogisticGradient.html" target="_top">No Frames</a></li>
</ul>
<ul class="navList" id="allclasses_navbar_top">
<li><a href="../../../../../allclasses-noframe.html">All Classes</a></li>
</ul>
<div>
<script type="text/javascript"><!--
allClassesLink = document.getElementById("allclasses_navbar_top");
if(window==top) {
allClassesLink.style.display = "block";
}
else {
allClassesLink.style.display = "none";
}
//-->
</script>
</div>
<div>
<ul class="subNavList">
<li>Summary: </li>
<li>Nested | </li>
<li>Field | </li>
<li><a href="#constructor_summary">Constr</a> | </li>
<li><a href="#method_summary">Method</a></li>
</ul>
<ul class="subNavList">
<li>Detail: </li>
<li>Field | </li>
<li><a href="#constructor_detail">Constr</a> | </li>
<li><a href="#method_detail">Method</a></li>
</ul>
</div>
<a name="skip-navbar_top">
<!-- -->
</a></div>
<!-- ========= END OF TOP NAVBAR ========= -->
<!-- ======== START OF CLASS DATA ======== -->
<div class="header">
<div class="subTitle">org.apache.spark.mllib.optimization</div>
<h2 title="Class LogisticGradient" class="title">Class LogisticGradient</h2>
</div>
<div class="contentContainer">
<ul class="inheritance">
<li>java.lang.Object</li>
<li>
<ul class="inheritance">
<li><a href="../../../../../org/apache/spark/mllib/optimization/Gradient.html" title="class in org.apache.spark.mllib.optimization">org.apache.spark.mllib.optimization.Gradient</a></li>
<li>
<ul class="inheritance">
<li>org.apache.spark.mllib.optimization.LogisticGradient</li>
</ul>
</li>
</ul>
</li>
</ul>
<div class="description">
<ul class="blockList">
<li class="blockList">
<dl>
<dt>All Implemented Interfaces:</dt>
<dd>java.io.Serializable</dd>
</dl>
<hr>
<br>
<pre>public class <span class="strong">LogisticGradient</span>
extends <a href="../../../../../org/apache/spark/mllib/optimization/Gradient.html" title="class in org.apache.spark.mllib.optimization">Gradient</a></pre>
<div class="block">:: DeveloperApi ::
Compute gradient and loss for a multinomial logistic loss function, as used
in multi-class classification (it is also used in binary logistic regression).
<p>
In <code>The Elements of Statistical Learning: Data Mining, Inference, and Prediction, 2nd Edition</code>
by Trevor Hastie, Robert Tibshirani, and Jerome Friedman, which can be downloaded from
http://statweb.stanford.edu/~tibs/ElemStatLearn/ , Eq. (4.17) on page 119 gives the formula of
multinomial logistic regression model. A simple calculation shows that
<p>
<pre><code>
P(y=0|x, w) = 1 / (1 + \sum_i^{K-1} \exp(x w_i))
P(y=1|x, w) = exp(x w_1) / (1 + \sum_i^{K-1} \exp(x w_i))
...
P(y=K-1|x, w) = exp(x w_{K-1}) / (1 + \sum_i^{K-1} \exp(x w_i))
</code></pre>
<p>
for K classes multiclass classification problem.
<p>
The model weights w = (w_1, w_2, ..., w_{K-1})^T becomes a matrix which has dimension of
(K-1) * (N+1) if the intercepts are added. If the intercepts are not added, the dimension
will be (K-1) * N.
<p>
As a result, the loss of objective function for a single instance of data can be written as
<pre><code>
l(w, x) = -log P(y|x, w) = -\alpha(y) log P(y=0|x, w) - (1-\alpha(y)) log P(y|x, w)
= log(1 + \sum_i^{K-1}\exp(x w_i)) - (1-\alpha(y)) x w_{y-1}
= log(1 + \sum_i^{K-1}\exp(margins_i)) - (1-\alpha(y)) margins_{y-1}
</code></pre>
<p>
where \alpha(i) = 1 if i != 0, and
\alpha(i) = 0 if i == 0,
margins_i = x w_i.
<p>
For optimization, we have to calculate the first derivative of the loss function, and
a simple calculation shows that
<p>
<pre><code>
\frac{\partial l(w, x)}{\partial w_{ij}}
= (\exp(x w_i) / (1 + \sum_k^{K-1} \exp(x w_k)) - (1-\alpha(y)\delta_{y, i+1})) * x_j
= multiplier_i * x_j
</code></pre>
<p>
where \delta_{i, j} = 1 if i == j,
\delta_{i, j} = 0 if i != j, and
multiplier =
\exp(margins_i) / (1 + \sum_k^{K-1} \exp(margins_i)) - (1-\alpha(y)\delta_{y, i+1})
<p>
If any of margins is larger than 709.78, the numerical computation of multiplier and loss
function will be suffered from arithmetic overflow. This issue occurs when there are outliers
in data which are far away from hyperplane, and this will cause the failing of training once
infinity / infinity is introduced. Note that this is only a concern when max(margins) > 0.
<p>
Fortunately, when max(margins) = maxMargin > 0, the loss function and the multiplier can be
easily rewritten into the following equivalent numerically stable formula.
<p>
<pre><code>
l(w, x) = log(1 + \sum_i^{K-1}\exp(margins_i)) - (1-\alpha(y)) margins_{y-1}
= log(\exp(-maxMargin) + \sum_i^{K-1}\exp(margins_i - maxMargin)) + maxMargin
- (1-\alpha(y)) margins_{y-1}
= log(1 + sum) + maxMargin - (1-\alpha(y)) margins_{y-1}
</code></pre>
<p>
where sum = \exp(-maxMargin) + \sum_i^{K-1}\exp(margins_i - maxMargin) - 1.
<p>
Note that each term, (margins_i - maxMargin) in \exp is smaller than zero; as a result,
overflow will not happen with this formula.
<p>
For multiplier, similar trick can be applied as the following,
<p>
<pre><code>
multiplier = \exp(margins_i) / (1 + \sum_k^{K-1} \exp(margins_i)) - (1-\alpha(y)\delta_{y, i+1})
= \exp(margins_i - maxMargin) / (1 + sum) - (1-\alpha(y)\delta_{y, i+1})
</code></pre>
<p>
where each term in \exp is also smaller than zero, so overflow is not a concern.
<p>
For the detailed mathematical derivation, see the reference at
http://www.slideshare.net/dbtsai/2014-0620-mlor-36132297
<p>
param: numClasses the number of possible outcomes for k classes classification problem in
Multinomial Logistic Regression. By default, it is binary logistic regression
so numClasses will be set to 2.</div>
<dl><dt><span class="strong">See Also:</span></dt><dd><a href="../../../../../serialized-form.html#org.apache.spark.mllib.optimization.LogisticGradient">Serialized Form</a></dd></dl>
</li>
</ul>
</div>
<div class="summary">
<ul class="blockList">
<li class="blockList">
<!-- ======== CONSTRUCTOR SUMMARY ======== -->
<ul class="blockList">
<li class="blockList"><a name="constructor_summary">
<!-- -->
</a>
<h3>Constructor Summary</h3>
<table class="overviewSummary" border="0" cellpadding="3" cellspacing="0" summary="Constructor Summary table, listing constructors, and an explanation">
<caption><span>Constructors</span><span class="tabEnd"> </span></caption>
<tr>
<th class="colOne" scope="col">Constructor and Description</th>
</tr>
<tr class="altColor">
<td class="colOne"><code><strong><a href="../../../../../org/apache/spark/mllib/optimization/LogisticGradient.html#LogisticGradient()">LogisticGradient</a></strong>()</code> </td>
</tr>
<tr class="rowColor">
<td class="colOne"><code><strong><a href="../../../../../org/apache/spark/mllib/optimization/LogisticGradient.html#LogisticGradient(int)">LogisticGradient</a></strong>(int numClasses)</code> </td>
</tr>
</table>
</li>
</ul>
<!-- ========== METHOD SUMMARY =========== -->
<ul class="blockList">
<li class="blockList"><a name="method_summary">
<!-- -->
</a>
<h3>Method Summary</h3>
<table class="overviewSummary" border="0" cellpadding="3" cellspacing="0" summary="Method Summary table, listing methods, and an explanation">
<caption><span>Methods</span><span class="tabEnd"> </span></caption>
<tr>
<th class="colFirst" scope="col">Modifier and Type</th>
<th class="colLast" scope="col">Method and Description</th>
</tr>
<tr class="altColor">
<td class="colFirst"><code>scala.Tuple2<<a href="../../../../../org/apache/spark/mllib/linalg/Vector.html" title="interface in org.apache.spark.mllib.linalg">Vector</a>,java.lang.Object></code></td>
<td class="colLast"><code><strong><a href="../../../../../org/apache/spark/mllib/optimization/LogisticGradient.html#compute(org.apache.spark.mllib.linalg.Vector, double, org.apache.spark.mllib.linalg.Vector)">compute</a></strong>(<a href="../../../../../org/apache/spark/mllib/linalg/Vector.html" title="interface in org.apache.spark.mllib.linalg">Vector</a> data,
double label,
<a href="../../../../../org/apache/spark/mllib/linalg/Vector.html" title="interface in org.apache.spark.mllib.linalg">Vector</a> weights)</code>
<div class="block">Compute the gradient and loss given the features of a single data point.</div>
</td>
</tr>
<tr class="rowColor">
<td class="colFirst"><code>double</code></td>
<td class="colLast"><code><strong><a href="../../../../../org/apache/spark/mllib/optimization/LogisticGradient.html#compute(org.apache.spark.mllib.linalg.Vector, double, org.apache.spark.mllib.linalg.Vector, org.apache.spark.mllib.linalg.Vector)">compute</a></strong>(<a href="../../../../../org/apache/spark/mllib/linalg/Vector.html" title="interface in org.apache.spark.mllib.linalg">Vector</a> data,
double label,
<a href="../../../../../org/apache/spark/mllib/linalg/Vector.html" title="interface in org.apache.spark.mllib.linalg">Vector</a> weights,
<a href="../../../../../org/apache/spark/mllib/linalg/Vector.html" title="interface in org.apache.spark.mllib.linalg">Vector</a> cumGradient)</code>
<div class="block">Compute the gradient and loss given the features of a single data point,
add the gradient to a provided vector to avoid creating new objects, and return loss.</div>
</td>
</tr>
</table>
<ul class="blockList">
<li class="blockList"><a name="methods_inherited_from_class_java.lang.Object">
<!-- -->
</a>
<h3>Methods inherited from class java.lang.Object</h3>
<code>clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait</code></li>
</ul>
</li>
</ul>
</li>
</ul>
</div>
<div class="details">
<ul class="blockList">
<li class="blockList">
<!-- ========= CONSTRUCTOR DETAIL ======== -->
<ul class="blockList">
<li class="blockList"><a name="constructor_detail">
<!-- -->
</a>
<h3>Constructor Detail</h3>
<a name="LogisticGradient(int)">
<!-- -->
</a>
<ul class="blockList">
<li class="blockList">
<h4>LogisticGradient</h4>
<pre>public LogisticGradient(int numClasses)</pre>
</li>
</ul>
<a name="LogisticGradient()">
<!-- -->
</a>
<ul class="blockListLast">
<li class="blockList">
<h4>LogisticGradient</h4>
<pre>public LogisticGradient()</pre>
</li>
</ul>
</li>
</ul>
<!-- ============ METHOD DETAIL ========== -->
<ul class="blockList">
<li class="blockList"><a name="method_detail">
<!-- -->
</a>
<h3>Method Detail</h3>
<a name="compute(org.apache.spark.mllib.linalg.Vector, double, org.apache.spark.mllib.linalg.Vector)">
<!-- -->
</a>
<ul class="blockList">
<li class="blockList">
<h4>compute</h4>
<pre>public scala.Tuple2<<a href="../../../../../org/apache/spark/mllib/linalg/Vector.html" title="interface in org.apache.spark.mllib.linalg">Vector</a>,java.lang.Object> compute(<a href="../../../../../org/apache/spark/mllib/linalg/Vector.html" title="interface in org.apache.spark.mllib.linalg">Vector</a> data,
double label,
<a href="../../../../../org/apache/spark/mllib/linalg/Vector.html" title="interface in org.apache.spark.mllib.linalg">Vector</a> weights)</pre>
<div class="block"><strong>Description copied from class: <code><a href="../../../../../org/apache/spark/mllib/optimization/Gradient.html#compute(org.apache.spark.mllib.linalg.Vector, double, org.apache.spark.mllib.linalg.Vector)">Gradient</a></code></strong></div>
<div class="block">Compute the gradient and loss given the features of a single data point.
<p></div>
<dl>
<dt><strong>Overrides:</strong></dt>
<dd><code><a href="../../../../../org/apache/spark/mllib/optimization/Gradient.html#compute(org.apache.spark.mllib.linalg.Vector, double, org.apache.spark.mllib.linalg.Vector)">compute</a></code> in class <code><a href="../../../../../org/apache/spark/mllib/optimization/Gradient.html" title="class in org.apache.spark.mllib.optimization">Gradient</a></code></dd>
<dt><span class="strong">Parameters:</span></dt><dd><code>data</code> - features for one data point</dd><dd><code>label</code> - label for this data point</dd><dd><code>weights</code> - weights/coefficients corresponding to features
<p></dd>
<dt><span class="strong">Returns:</span></dt><dd>(gradient: Vector, loss: Double)</dd></dl>
</li>
</ul>
<a name="compute(org.apache.spark.mllib.linalg.Vector, double, org.apache.spark.mllib.linalg.Vector, org.apache.spark.mllib.linalg.Vector)">
<!-- -->
</a>
<ul class="blockListLast">
<li class="blockList">
<h4>compute</h4>
<pre>public double compute(<a href="../../../../../org/apache/spark/mllib/linalg/Vector.html" title="interface in org.apache.spark.mllib.linalg">Vector</a> data,
double label,
<a href="../../../../../org/apache/spark/mllib/linalg/Vector.html" title="interface in org.apache.spark.mllib.linalg">Vector</a> weights,
<a href="../../../../../org/apache/spark/mllib/linalg/Vector.html" title="interface in org.apache.spark.mllib.linalg">Vector</a> cumGradient)</pre>
<div class="block"><strong>Description copied from class: <code><a href="../../../../../org/apache/spark/mllib/optimization/Gradient.html#compute(org.apache.spark.mllib.linalg.Vector, double, org.apache.spark.mllib.linalg.Vector, org.apache.spark.mllib.linalg.Vector)">Gradient</a></code></strong></div>
<div class="block">Compute the gradient and loss given the features of a single data point,
add the gradient to a provided vector to avoid creating new objects, and return loss.
<p></div>
<dl>
<dt><strong>Specified by:</strong></dt>
<dd><code><a href="../../../../../org/apache/spark/mllib/optimization/Gradient.html#compute(org.apache.spark.mllib.linalg.Vector, double, org.apache.spark.mllib.linalg.Vector, org.apache.spark.mllib.linalg.Vector)">compute</a></code> in class <code><a href="../../../../../org/apache/spark/mllib/optimization/Gradient.html" title="class in org.apache.spark.mllib.optimization">Gradient</a></code></dd>
<dt><span class="strong">Parameters:</span></dt><dd><code>data</code> - features for one data point</dd><dd><code>label</code> - label for this data point</dd><dd><code>weights</code> - weights/coefficients corresponding to features</dd><dd><code>cumGradient</code> - the computed gradient will be added to this vector
<p></dd>
<dt><span class="strong">Returns:</span></dt><dd>loss</dd></dl>
</li>
</ul>
</li>
</ul>
</li>
</ul>
</div>
</div>
<!-- ========= END OF CLASS DATA ========= -->
<!-- ======= START OF BOTTOM NAVBAR ====== -->
<div class="bottomNav"><a name="navbar_bottom">
<!-- -->
</a><a href="#skip-navbar_bottom" title="Skip navigation links"></a><a name="navbar_bottom_firstrow">
<!-- -->
</a>
<ul class="navList" title="Navigation">
<li><a href="../../../../../overview-summary.html">Overview</a></li>
<li><a href="package-summary.html">Package</a></li>
<li class="navBarCell1Rev">Class</li>
<li><a href="package-tree.html">Tree</a></li>
<li><a href="../../../../../deprecated-list.html">Deprecated</a></li>
<li><a href="../../../../../index-all.html">Index</a></li>
<li><a href="../../../../../help-doc.html">Help</a></li>
</ul>
</div>
<div class="subNav">
<ul class="navList">
<li><a href="../../../../../org/apache/spark/mllib/optimization/LeastSquaresGradient.html" title="class in org.apache.spark.mllib.optimization"><span class="strong">Prev Class</span></a></li>
<li><a href="../../../../../org/apache/spark/mllib/optimization/Optimizer.html" title="interface in org.apache.spark.mllib.optimization"><span class="strong">Next Class</span></a></li>
</ul>
<ul class="navList">
<li><a href="../../../../../index.html?org/apache/spark/mllib/optimization/LogisticGradient.html" target="_top">Frames</a></li>
<li><a href="LogisticGradient.html" target="_top">No Frames</a></li>
</ul>
<ul class="navList" id="allclasses_navbar_bottom">
<li><a href="../../../../../allclasses-noframe.html">All Classes</a></li>
</ul>
<div>
<script type="text/javascript"><!--
allClassesLink = document.getElementById("allclasses_navbar_bottom");
if(window==top) {
allClassesLink.style.display = "block";
}
else {
allClassesLink.style.display = "none";
}
//-->
</script>
</div>
<div>
<ul class="subNavList">
<li>Summary: </li>
<li>Nested | </li>
<li>Field | </li>
<li><a href="#constructor_summary">Constr</a> | </li>
<li><a href="#method_summary">Method</a></li>
</ul>
<ul class="subNavList">
<li>Detail: </li>
<li>Field | </li>
<li><a href="#constructor_detail">Constr</a> | </li>
<li><a href="#method_detail">Method</a></li>
</ul>
</div>
<a name="skip-navbar_bottom">
<!-- -->
</a></div>
<!-- ======== END OF BOTTOM NAVBAR ======= -->
<script defer="defer" type="text/javascript" src="../../../../../lib/jquery.js"></script><script defer="defer" type="text/javascript" src="../../../../../lib/api-javadocs.js"></script></body>
</html>
|
