3 import static org.testng.Assert.assertEquals;
4 import static org.testng.Assert.assertFalse;
5 import static org.testng.Assert.assertTrue;
6 import static org.testng.Assert.fail;
8 import java.util.Random;
10 import org.testng.annotations.Test;
11 import org.testng.internal.junit.ArrayAsserts;
13 public class SparseMatrixTest
15 final static double DELTA = 0.0001d;
17 Random r = new Random(1729);
19 @Test(groups = "Functional")
20 public void testConstructor()
22 MatrixI m1 = new SparseMatrix(
23 new double[][] { { 2, 0, 4 }, { 0, 6, 0 } });
24 assertEquals(m1.getValue(0, 0), 2d);
25 assertEquals(m1.getValue(0, 1), 0d);
26 assertEquals(m1.getValue(0, 2), 4d);
27 assertEquals(m1.getValue(1, 0), 0d);
28 assertEquals(m1.getValue(1, 1), 6d);
29 assertEquals(m1.getValue(1, 2), 0d);
32 @Test(groups = "Functional")
33 public void testTranspose()
35 MatrixI m1 = new SparseMatrix(
36 new double[][] { { 2, 0, 4 }, { 5, 6, 0 } });
37 MatrixI m2 = m1.transpose();
38 assertTrue(m2 instanceof SparseMatrix);
39 assertEquals(m2.height(), 3);
40 assertEquals(m2.width(), 2);
41 assertEquals(m2.getValue(0, 0), 2d);
42 assertEquals(m2.getValue(0, 1), 5d);
43 assertEquals(m2.getValue(1, 0), 0d);
44 assertEquals(m2.getValue(1, 1), 6d);
45 assertEquals(m2.getValue(2, 0), 4d);
46 assertEquals(m2.getValue(2, 1), 0d);
48 @Test(groups = "Functional")
49 public void testPreMultiply()
51 MatrixI m1 = new SparseMatrix(new double[][] { { 2, 3, 4 } }); // 1x3
52 MatrixI m2 = new SparseMatrix(new double[][] { { 5 }, { 6 }, { 7 } }); // 3x1
55 * 1x3 times 3x1 is 1x1
56 * 2x5 + 3x6 + 4*7 = 56
58 MatrixI m3 = m2.preMultiply(m1);
59 assertFalse(m3 instanceof SparseMatrix);
60 assertEquals(m3.height(), 1);
61 assertEquals(m3.width(), 1);
62 assertEquals(m3.getValue(0, 0), 56d);
65 * 3x1 times 1x3 is 3x3
67 m3 = m1.preMultiply(m2);
68 assertEquals(m3.height(), 3);
69 assertEquals(m3.width(), 3);
70 assertEquals(m3.getValue(0, 0), 10d);
71 assertEquals(m3.getValue(0, 1), 15d);
72 assertEquals(m3.getValue(0, 2), 20d);
73 assertEquals(m3.getValue(1, 0), 12d);
74 assertEquals(m3.getValue(1, 1), 18d);
75 assertEquals(m3.getValue(1, 2), 24d);
76 assertEquals(m3.getValue(2, 0), 14d);
77 assertEquals(m3.getValue(2, 1), 21d);
78 assertEquals(m3.getValue(2, 2), 28d);
82 groups = "Functional",
83 expectedExceptions = { IllegalArgumentException.class })
84 public void testPreMultiply_tooManyColumns()
86 Matrix m1 = new SparseMatrix(
87 new double[][] { { 2, 3, 4 }, { 3, 4, 5 } }); // 2x3
90 * 2x3 times 2x3 invalid operation -
91 * multiplier has more columns than multiplicand has rows
94 fail("Expected exception");
98 groups = "Functional",
99 expectedExceptions = { IllegalArgumentException.class })
100 public void testPreMultiply_tooFewColumns()
102 Matrix m1 = new SparseMatrix(
103 new double[][] { { 2, 3, 4 }, { 3, 4, 5 } }); // 2x3
106 * 3x2 times 3x2 invalid operation -
107 * multiplier has more columns than multiplicand has row
110 fail("Expected exception");
113 @Test(groups = "Functional")
114 public void testPostMultiply()
124 MatrixI m1 = new SparseMatrix(new double[][] { { 2, 3 }, { 4, 5 } });
125 MatrixI m2 = new SparseMatrix(new double[][] { { 10, 100 },
127 MatrixI m3 = m1.postMultiply(m2);
128 assertEquals(m3.getValue(0, 0), 3020d);
129 assertEquals(m3.getValue(0, 1), 30200d);
130 assertEquals(m3.getValue(1, 0), 5040d);
131 assertEquals(m3.getValue(1, 1), 50400d);
134 * also check m2.preMultiply(m1) - should be same as m1.postMultiply(m2)
136 MatrixI m4 = m2.preMultiply(m1);
137 assertMatricesMatch(m3, m4, 0.00001d);
140 * m1 has more rows than columns
141 * (2).(10 100 1000) = (20 200 2000)
144 m1 = new SparseMatrix(new double[][] { { 2 }, { 3 } });
145 m2 = new SparseMatrix(new double[][] { { 10, 100, 1000 } });
146 m3 = m1.postMultiply(m2);
147 assertEquals(m3.height(), 2);
148 assertEquals(m3.width(), 3);
149 assertEquals(m3.getValue(0, 0), 20d);
150 assertEquals(m3.getValue(0, 1), 200d);
151 assertEquals(m3.getValue(0, 2), 2000d);
152 assertEquals(m3.getValue(1, 0), 30d);
153 assertEquals(m3.getValue(1, 1), 300d);
154 assertEquals(m3.getValue(1, 2), 3000d);
156 m4 = m2.preMultiply(m1);
157 assertMatricesMatch(m3, m4, 0.00001d);
160 * m1 has more columns than rows
161 * (2 3 4) . (5 4) = (56 25)
164 * [0, 0] = 2*5 + 3*6 + 4*7 = 56
165 * [0, 1] = 2*4 + 3*3 + 4*2 = 25
167 m1 = new SparseMatrix(new double[][] { { 2, 3, 4 } });
168 m2 = new SparseMatrix(new double[][] { { 5, 4 }, { 6, 3 }, { 7, 2 } });
169 m3 = m1.postMultiply(m2);
170 assertEquals(m3.height(), 1);
171 assertEquals(m3.width(), 2);
172 assertEquals(m3.getValue(0, 0), 56d);
173 assertEquals(m3.getValue(0, 1), 25d);
176 * and check premultiply equivalent
178 m4 = m2.preMultiply(m1);
179 assertMatricesMatch(m3, m4, 0.00001d);
182 @Test(groups = "Timing")
183 public void testSign()
185 assertEquals(Matrix.sign(-1, -2), -1d);
186 assertEquals(Matrix.sign(-1, 2), 1d);
187 assertEquals(Matrix.sign(-1, 0), 1d);
188 assertEquals(Matrix.sign(1, -2), -1d);
189 assertEquals(Matrix.sign(1, 2), 1d);
190 assertEquals(Matrix.sign(1, 0), 1d);
194 * Verify that the results of method tred() are the same for SparseMatrix as
195 * they are for Matrix (i.e. a regression test rather than an absolute test of
196 * correctness of results)
198 @Test(groups = "Functional")
199 public void testTred_matchesMatrix()
202 * make a pseudo-random symmetric matrix as required for tred/tqli
206 double[][] d = getSparseValues(rows, cols, 3);
209 * make a copy of the values so m1, m2 are not
212 double[][] d1 = new double[rows][cols];
213 for (int row = 0; row < rows; row++)
215 for (int col = 0; col < cols; col++)
217 d1[row][col] = d[row][col];
220 Matrix m1 = new Matrix(d);
221 Matrix m2 = new SparseMatrix(d1);
222 assertMatricesMatch(m1, m2, 0.00001d); // sanity check
225 assertMatricesMatch(m1, m2, 0.00001d);
228 private void assertMatricesMatch(MatrixI m1, MatrixI m2, double delta)
230 if (m1.height() != m2.height())
232 fail("height mismatch");
234 if (m1.width() != m2.width())
236 fail("width mismatch");
238 for (int row = 0; row < m1.height(); row++)
240 for (int col = 0; col < m1.width(); col++)
242 double v2 = m2.getValue(row, col);
243 double v1 = m1.getValue(row, col);
244 if (Math.abs(v1 - v2) > DELTA)
246 fail(String.format("At [%d, %d] %f != %f", row, col, v1, v2));
250 ArrayAsserts.assertArrayEquals(m1.getD(), m2.getD(), delta);
251 ArrayAsserts.assertArrayEquals(m1.getE(), m2.getE(), 0.00001d);
255 public void testGetValue()
257 double[][] d = new double[][] { { 0, 0, 1, 0, 0 }, { 2, 3, 0, 0, 0 },
259 MatrixI m = new SparseMatrix(d);
260 for (int row = 0; row < 3; row++)
262 for (int col = 0; col < 5; col++)
264 assertEquals(m.getValue(row, col), d[row][col],
265 String.format("At [%d, %d]", row, col));
271 * Verify that the results of method tqli() are the same for SparseMatrix as
272 * they are for Matrix (i.e. a regression test rather than an absolute test of
273 * correctness of results)
277 @Test(groups = "Functional")
278 public void testTqli_matchesMatrix() throws Exception
281 * make a pseudo-random symmetric matrix as required for tred
285 double[][] d = getSparseValues(rows, cols, 3);
288 * make a copy of the values so m1, m2 are not
291 double[][] d1 = new double[rows][cols];
292 for (int row = 0; row < rows; row++)
294 for (int col = 0; col < cols; col++)
296 d1[row][col] = d[row][col];
299 Matrix m1 = new Matrix(d);
300 Matrix m2 = new SparseMatrix(d1);
302 // have to do tred() before doing tqli()
305 assertMatricesMatch(m1, m2, 0.00001d);
309 assertMatricesMatch(m1, m2, 0.00001d);
313 * Helper method to make values for a sparse, pseudo-random symmetric matrix
318 * one in 'occupancy' entries will be non-zero
321 public double[][] getSparseValues(int rows, int cols, int occupancy)
324 * generate whole number values between -12 and +12
325 * (to mimic score matrices used in Jalview)
327 double[][] d = new double[rows][cols];
329 for (int i = 0; i < rows; i++)
331 if (++m % occupancy == 0)
333 d[i][i] = r.nextInt() % 13; // diagonal
335 for (int j = 0; j < i; j++)
337 if (++m % occupancy == 0)
339 d[i][j] = r.nextInt() % 13;
349 * Test that verifies that the result of preMultiply is a SparseMatrix if more
350 * than 80% zeroes, else a Matrix
352 @Test(groups = "Functional")
353 public void testPreMultiply_sparseProduct()
355 MatrixI m1 = new SparseMatrix(new double[][] { { 1 }, { 0 }, { 0 },
356 { 0 }, { 0 } }); // 5x1
357 MatrixI m2 = new SparseMatrix(new double[][] { { 1, 1, 1, 1 } }); // 1x4
360 * m1.m2 makes a row of 4 1's, and 4 rows of zeros
361 * 20% non-zero so not 'sparse'
363 MatrixI m3 = m2.preMultiply(m1);
364 assertFalse(m3 instanceof SparseMatrix);
367 * replace a 1 with a 0 in the product:
368 * it is now > 80% zero so 'sparse'
370 m2 = new SparseMatrix(new double[][] { { 1, 1, 1, 0 } });
371 m3 = m2.preMultiply(m1);
372 assertTrue(m3 instanceof SparseMatrix);
375 @Test(groups = "Functional")
376 public void testFillRatio()
378 SparseMatrix m1 = new SparseMatrix(new double[][] { { 2, 0, 4, 1, 0 },
379 { 0, 6, 0, 0, 0 } });
380 assertEquals(m1.getFillRatio(), 0.4f);
384 * Verify that the results of method tred() are the same if the calculation is
387 @Test(groups = "Functional")
388 public void testTred_reproducible()
391 * make a pseudo-random symmetric matrix as required for tred/tqli
395 double[][] d = getSparseValues(rows, cols, 3);
398 * make a copy of the values so m1, m2 are not
401 double[][] d1 = new double[rows][cols];
402 for (int row = 0; row < rows; row++)
404 for (int col = 0; col < cols; col++)
406 d1[row][col] = d[row][col];
409 Matrix m1 = new SparseMatrix(d);
410 Matrix m2 = new SparseMatrix(d1);
411 assertMatricesMatch(m1, m2, 1.0e16); // sanity check
414 assertMatricesMatch(m1, m2, 0.00001d);