forked from libigl/libigl
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathmatlab-to-eigen.html
271 lines (230 loc) · 10.2 KB
/
matlab-to-eigen.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
<html> <head>
<title>MATLAB to Eigen</title>
<link href="./style.css" rel="stylesheet" type="text/css">
</head>
<body>
<table>
<tr class="header">
<th>MATLAB</th>
<th>Eigen with libigl</th>
<th>Notes</th>
</tr>
<tr class=d0>
<td><pre><code>[Y,IX] = sort(Y,dim,mode)</code></pre></td>
<td><pre><code>igl::sort(X,dim,mode,Y,IX)</code></pre></td>
<td>MATLAB version allows Y to be a multidimensional matrix, but the
Eigen version is only for 1D or 2D matrices.</td>
</tr>
<tr class=d1>
<td><pre><code>B(i:(i+w),j:(j+h)) = A(x:(x+w),y:(y+h))</code></pre></td>
<td><pre><code>B.block(i,j,w,h) = A.block(i,j,w,h)</code></pre></td>
<td>MATLAB version would allow w and h to be non-positive since the
colon operator evaluates to a list of indices, but the Eigen version
needs non-negative width and height values.</td>
</tr>
<tr class=d0>
<td><pre><code>max(A(:))</code></pre></td>
<td><pre><code>A.maxCoeff()</code></pre></td>
<td>Find the maximum coefficient over all entries of the matrix.</td>
</tr>
<tr class=d1>
<td><pre><code>min(A(:))</code></pre></td>
<td><pre><code>A.minCoeff()</code></pre></td>
<td>Find the minimum coefficient over all entries of the matrix.</td>
</tr>
<tr class=d0>
<td><pre><code>eye(w,h)</code></pre></td>
<td><pre><code>MatrixXd::Identity(w,h), MatrixXf::Identity(w,h), etc.</code></pre></td>
<td></td>
</tr>
<tr class=d1>
<td><pre><code>A(i:(i+w),j:(j+h)) = eye(w,h)</code></pre></td>
<td><pre><code>A.setIdentity()</code></pre></td>
<td></td>
</tr>
<tr class=d0>
<td><pre><code>[I,J,V] = find(X)</code></pre></td>
<td><pre><code>igl::find(X,I,J,V)</code></pre></td>
<td>Matlab supports finding subscripts (I and J) as well as indices
(just I), but so far igl::find only supports subscripts. Also,
igl::find requires X to be sparse.</td>
</tr>
<tr class=d1>
<td><pre><code>X(:,j) = X(:,j) + x</code></pre></td>
<td><pre><code>X.col(j).array() += x</code></pre></td>
<td></td>
</tr>
<tr class=d0>
<td><pre><code>Acol_sum = sum(A,1)<br>Arow_sum = sum(A,2)<br>Adim_sum = sum(Asparse,dim)</code></pre></td>
<td><pre><code>Acol_sum = A.colwise().sum()<br>Arow_sum = A.rowwise().sum()<br>igl::sum(Asparse,dim,Adim_sum)</code></pre></td>
<td>Currently the igl version only supports sparse matrix input (and
dim must be 1 or 2)</td>
</tr>
<tr class=d1>
<td><pre><code>D = diag(M)</code></pre></td>
<td><pre><code>igl::diag(M,D)</code></pre></td>
<td>Extract the main diagonal of a matrix. Currently igl version
supports sparse only.</td>
</tr>
<tr class=d0>
<td><pre><code>M = diag(D)</code></pre></td>
<td><pre><code>igl::diag(D,M)</code></pre></td>
<td>Construct new square matrix M with entries of vector D along the
diagonal. Currently igl version supports sparse only.</td>
</tr>
<tr class=d1>
<td><pre><code>[Y,I] = max(X,[],dim)</code></pre></td>
<td id=mat_max><pre><code>igl::mat_max(X,dim,Y,I)</code></pre></td>
<td>Matlab has a bizarre convention of passing [] as the second
argument to mean take the max/min along dimension dim.</td>
</tr>
<tr class=d0>
<td><pre><code>Y = max(X,[],1)<br>Y = max(X,[],2)<br>Y = min(X,[],1)<br>Y = min(X,[],2)</code></pre></td>
<td><pre><code>Y = X.colwise().maxCoeff()<br>Y = X.rowwise().maxCoeff()<br>Y = X.colwise().minCoeff()<br>Y = X.rowwise().minCoeff()</code></pre></td>
<td>Matlab allows you to obtain the indices of extrema see <a href=#mat_max>mat_max</a></td>
</tr>
<tr class=d1>
<td><pre><code>C = A.*B</code></pre></td>
<td><pre><code>C = (A.array() * B.array()).matrix()</code></pre></td>
<td></td>
</tr>
<tr class=d0>
<td><pre><code>C = A.^b</code></pre></td>
<td><pre><code>C = A.array().pow(b).matrix()</code></pre></td>
<td></td>
</tr>
<tr class=d1>
<td><pre><code>A(B == 0) = C(B==0)</code></pre></td>
<td><pre><code>A = (B.array() == 0).select(C,A)</code></pre></td>
<td></td>
</tr>
<tr class=gotcha1>
<td><pre><code>C = A + B'</code></pre></td>
<td><pre><code>SparseMatrixType BT = B.transpose()<br>SparseMatrixType C = A+BT;</code></pre></td>
<td>Do <b>not</b> attempt to combine .transpose() in expression like
this: <pre><code>C = A + B.transpose()</code></pre></td>
</tr>
<tr class=gotcha2>
<td><pre><code>[L,p] = chol(A)</code></pre></td>
<td><pre><code>SparseLLT<SparseMatrixType> A_LLT(A.template triangularView<Lower>())<br>SparseMatrixType L = A_LLT.matrixL();bool p = (L*0).eval().nonZeros()==0;</code></pre></td>
<td>Do <b>not</b> attempt to use A in constructor of A_LLT like
this: <pre><code>SparseLLT<SparseMatrixType> A_LLT(A)</code></pre><br>
Do <b>not</b> attempt to use A_LLT.succeeded() to determine if Cholesky
factorization succeeded, like this:
<pre><code>bool p = A_LLT.succeeded()</code></pre>
</td>
</tr>
<tr class=d0>
<td><pre><code>X = U\(L\b)</code></pre></td>
<td><pre><code>X = b;<br>L.template triangularView<Lower>().solveInPlace(X);<br>U.template triangularView<Upper>().solveInPlace(X);</code></pre></td>
<td>Expects that L and U are lower and upper triangular matrices
respectively</td>
</tr>
<tr class=d1>
<td><pre><code>B = repmat(A,i,j)</code></pre></td>
<td><pre><code>igl::repmat(A,i,j,B)</code></pre></td>
<td>So far igl::repmat is only implemented for dense matrices.
</td>
</tr>
<tr class=d0>
<td><pre><code>I = low:step:hi</code></pre></td>
<td><pre><code>igl::colon(low,step,hi,I)</code></pre></td>
<td>IGL version should be templated enough to handle same situations as
matlab's colon. The matlab keyword <b>end</b> does not correspond in
the C++ version. You'll have to use M.size(),M.rows() or whatever.
</td>
</tr>
<tr class=d1>
<td><pre><code>O = ones(m,n)</code></pre></td>
<td><pre><code>Matrix* O = Matrix*::Ones(m,n)</code></pre></td>
<td></td>
</tr>
<tr class=d0>
<td><pre><code>O = zeros(m,n)</code></pre></td>
<td><pre><code>Matrix* O = Matrix*::Zero(m,n)</code></pre></td>
<td></td>
</tr>
<tr class=d1>
<td><pre><code>B = A(I,J)<br>B = A(I,:)</code></pre></td>
<td><pre><code>igl::slice(A,I,J,B)<br>B = igl::slice(A,I,igl::colon(0,A.cols()-1))</code></pre></td>
<td>This is only for the case when I and J are lists of indices and
not vectors of logicals.</td>
</tr>
<tr class=d0>
<td><pre><code>B(I,J) = A<br>B(I,:) = A</code></pre></td>
<td><pre><code>igl::slice_into(A,I,J,B)<br>B = igl::slice_into(A,I,igl::colon(0,B.cols()-1))</code></pre></td>
<td>This is only for the case when I and J are lists of indices and
not vectors of logicals.</td>
</tr>
<tr class=d1>
<td><pre><code>M = mode(X,dim)</code></pre></td>
<td><pre><code>igl::mode(X,dim,M)</code></pre></td>
<td></td>
</tr>
<tr class=d0>
<td><pre><code>B = arrayfun(FUN, A)</code></pre></td>
<td><pre><code>B = A.unaryExpr(ptr_fun(FUN))</code></pre></td>
<td>If FUN is templated, the templates must be fully resolved.</td>
</tr>
<tr class=d1>
<td><pre><code>B = fliplr(A)<br>B = flipud(A)</code></pre></td>
<td><pre><code>B = A.rowwise().reverse().eval()<br>B =
A.colwise().reverse().eval()</code></pre></td>
<td>The <code>.eval()</code> is not necessary if A != B</td>
</tr>
<tr class=d0>
<td><pre><code>B = IM(A)
A = IM(A);
</code></pre></td>
<td><pre><code>B = A.unaryExpr(bind1st(mem_fun(
static_cast<VectorXi::Scalar&(VectorXi::*)(VectorXi::Index)>
(&VectorXi::operator())), &IM)).eval();
for_each(A.data(),A.data()+A.size(),[&IM](int & a){a=IM(a);});
</code></pre></td>
<td>Where IM is an "index map" column vector and A is an arbitrary
matrix. The <code>.eval()</code> is not necessary if A != B, but then
the second option should be used.</td>
</tr>
<tr class=d1>
<td><pre><code>A = sparse(I,J,V)</code></pre></td>
<td><pre><code>// build std::vector<Eigen::Triplet> IJV
A.setFromTriplets(IJV);
</code></pre></td>
<td>IJV and A should not be empty! (this might be fixed)</td>
</tr>
<tr class=d0>
<td><pre><code>A = min(A,c);</code></pre></td>
<td><pre><code>C.array() = A.array().min(c);
</code></pre></td>
<td>Coefficient-wise minimum of matrix and scalar (or matching size matrix)</td>
</tr>
<tr class=d1>
<td><pre><code>IP = I(P==0);</code></pre></td>
<td><pre><code>VectorXi IP = I;
IP.conservativeResize(stable_partition(
IP.data(),
IP.data()+IP.size(),
[&P](int i){return P(i)==0;})-IP.data());
</code></pre></td>
<td>Where I and P are vectors. <i>Requires C++11 and <code>#include <algorithm></code></i></td>
</tr>
<tr class=d0>
<td><pre><code>a = any(A(:))</code></pre></td>
<td><pre><code>bool a = any_of(A.data(),A.data()+A.size(),[](bool a){ return a;});</code></pre></td>
<td>Casts <code>Matrix::Scalar<code> to <code>bool</code>.</td>
</tr>
<!-- Insert rows for each command pair -->
<!-- Leave this here for copy and pasting
<tr class=d0>
<td><pre><code>Matlab code</code></pre></td>
<td><pre><code>Eigen code</code></pre></td>
<td>Notes</td>
</tr>
-->
</table>
<a href="http://eigen.tuxfamily.org/dox/AsciiQuickReference.txt">Eigen's
"ASCII Quick Reference" with MATLAB translations</a>
<br>
<a href="./tutorial.html">IGL Lib Tutorial</a>
</body>
</html>