mvpa.tests.test_metadataset

30

31 - def testSimple(self):

32 # bunch of datasets 33 34 datasets = [ 35 Dataset(samples=N.arange(12).reshape((4,3)), labels=1), 36 Dataset(samples=N.zeros((4,4)), labels=1), 37 Dataset(samples=N.ones((4,2), dtype='float'), labels=1), 38 ] 39 40 mds = MetaDataset(datasets) 41 42 # all together 43 self.failUnless(mds.samples.shape == (4, 9)) 44 # should do upcasting 45 self.failUnless(mds.samples.dtype == 'float') 46 # simple samples attrs 47 self.failUnless((mds.labels == [1] * 4).all()) 48 self.failUnless((mds.chunks == range(4)).all()) 49 50 # do sample selection across all datasets 51 mds1 = mds.selectSamples([0,3]) 52 self.failUnless(mds1.samples.shape == (2, 9)) 53 self.failUnless(\ 54 (mds1.samples[0] == [0, 1, 2, 0, 0, 0, 0, 1, 1]).all()) 55 self.failUnless(\ 56 (mds1.samples[1] == [9, 10, 11, 0, 0, 0, 0, 1, 1]).all()) 57 58 # more tricky feature selection on all datasets 59 mds2 = mds.selectFeatures([1,4,8]) 60 61 self.failUnless(\ 62 (mds2.samples == [[ 1, 0, 1], 63 [ 4, 0, 1], 64 [ 7, 0, 1], 65 [10, 0, 1]] ).all())

66 67

68 - def testMapping(self):

69 if not externals.exists('nifti'): 70 return 71 from mvpa.datasets.nifti import NiftiDataset 72 eeds = EEPDataset(os.path.join(pymvpa_dataroot, 'eep.bin'), labels=[1,2]) 73 nids = NiftiDataset(os.path.join(pymvpa_dataroot, 'example4d.nii.gz'), 74 labels=[1,2]) 75 plainds = Dataset(samples=N.arange(8).reshape((2,4)), labels=[1,2]) 76 77 datasets = (eeds, plainds, nids) 78 79 mds = MetaDataset(datasets) 80 81 self.failUnless(mds.nfeatures == N.sum([d.nfeatures for d in datasets])) 82 self.failUnless(mds.nsamples == 2) 83 84 # try reverse mapping 85 mr = mds.mapReverse(N.arange(mds.nfeatures)) 86 87 self.failUnless(len(mr) == 3) 88 self.failUnless(mr[1].shape == (plainds.nfeatures,))

89 90

91 - def testCombinedMapper(self):

92 # simple case: two array of different shape combined 93 m = CombinedMapper([DenseArrayMapper(mask=N.ones((2,3,4))), 94 MaskMapper(mask=N.array((1,1)))]) 95 96 self.failUnless(m.getInSize() == 26) 97 self.failUnless(m.getOutSize() == 26) 98 99 d1 = N.ones((5,2,3,4)) 100 d2_broken = N.ones((6,2)) + 1 101 d2 = N.ones((5,2)) + 1 102 103 # should not work for sample mismatch 104 self.failUnlessRaises(ValueError, m.forward, (d1, d2_broken)) 105 106 # check forward mapping (size and identity) 107 mf = m.forward((d1, d2)) 108 self.failUnless(mf.shape == (5, 26)) 109 self.failUnless((mf[:,:24] == 1).all()) 110 self.failUnless((mf[:,-2:] == 2).all()) 111 112 # check reverse mapping 113 self.failUnlessRaises(ValueError, m.reverse, N.arange(12)) 114 mr = m.reverse(N.arange(26) + 1) 115 self.failUnless(len(mr) == 2) 116 self.failUnless((mr[0] == N.arange(24).reshape((2,3,4)) + 1).all()) 117 self.failUnless((mr[1] == N.array((25,26))).all()) 118 119 # check reverse mapping of multiple samples 120 mr = m.reverse(N.array([N.arange(26) + 1 for i in range(4)])) 121 self.failUnless(len(mr) == 2) 122 self.failUnless( 123 (mr[0] == N.array([N.arange(24).reshape((2,3,4)) + 1 124 for i in range(4)])).all()) 125 self.failUnless( 126 (mr[1] == N.array([N.array((25,26)) for i in range(4)])).all()) 127 128 129 # check dummy train 130 m.train(Dataset(samples=N.random.rand(10,26), labels=range(10))) 131 self.failUnlessRaises(ValueError, m.train, 132 Dataset(samples=N.random.rand(10,25), labels=range(10))) 133 134 # check neighbor information 135 # fail if invalid id 136 self.failUnlessRaises(ValueError, m.getNeighbor, 26) 137 # neighbors for last feature of first mapper, ie. 138 # close in out space but infinite/undefined distance in in-space 139 self.failUnless([n for n in m.getNeighbor(23, radius=2)] 140 == [6, 7, 10, 11, 15, 18, 19, 21, 22, 23]) 141 142 # check feature selection 143 m.selectOut((23,25)) 144 self.failUnless(m.getInSize() == 26) 145 self.failUnless(m.getOutSize() == 2) 146 147 # check reverse mapping of truncated mapper 148 mr = m.reverse(N.array((99,88))) 149 target1 = N.zeros((2,3,4)) 150 target1[1,2,3] = 99 151 target2 = N.array((0, 88)) 152 self.failUnless(len(mr) == 2) 153 self.failUnless((mr[0] == target1).all()) 154 self.failUnless((mr[1] == target2).all()) 155 156 # check forward mapping 157 self.failUnless((m.forward((d1, d2))[0] == (1, 2)).all()) 158 159 # check copying 160 mc = deepcopy(m) 161 mc.selectOut([1]) 162 self.failUnless(m.getOutSize() == 2) 163 self.failUnless(mc.getOutSize() == 1)

164 165

166 - def testChainMapper(self):

167 data = N.array([N.arange(24).reshape(3,4,2) + (i * 100) 168 for i in range(10)]) 169 170 startpoints = [ 2, 4, 3, 5 ] 171 m = ChainMapper([BoxcarMapper(startpoints, 2), 172 DenseArrayMapper(mask=N.ones((2, 3, 4, 2)))]) 173 mp = m.forward(data) 174 # 4 startpoint, with each two samples of shape (3,4,2) 175 self.failUnless(mp.shape == (4, 48)) 176 177 self.failUnless(m.reverse(N.arange(48)).shape == (2, 3, 4, 2)) 178 179 # should behave a DenseArrayMapper alone 180 self.failUnless((N.array([n for n in m.getNeighbor(24, radius=1.1)]) 181 == N.array((0, 24, 25, 26, 32))).all())

182 183

184 - def testEventDataset(self):

185 # baisc checks 186 self.failUnlessRaises(DatasetError, EventDataset) 187 188 # simple data 189 samples = N.arange(240).reshape(10, 2, 3, 4) 190 191 # copy constructor does not work on non-2D data 192 self.failUnlessRaises(DatasetError, EventDataset, samples=samples) 193 194 # try case without extra features 195 evs = [Event(onset=2, duration=2, label=1, chunk=2), 196 Event(onset=5, duration=1, label=2, chunk=2), 197 Event(onset=7, duration=2, label=3, chunk=4)] 198 199 ds = EventDataset(samples=samples, events=evs) 200 self.failUnless(ds.nfeatures == 48) 201 self.failUnless(ds.nsamples == 3) 202 self.failUnless((ds.labels == [1,2,3]).all()) 203 self.failUnless((ds.chunks == [2,2,4]).all()) 204 mr = ds.mapReverse(N.arange(48)) 205 self.failUnless((mr == N.arange(48).reshape(2,2,3,4)).all()) 206 207 # try case with extra features 208 evs = [Event(onset=2, duration=2, label=1, features=[2,3]), 209 Event(onset=5, duration=2, label=1, features=[4,5]), 210 Event(onset=7, duration=2, label=1, features=[6,7]),] 211 ds = EventDataset(samples=samples, events=evs) 212 # we have 2 additional features 213 self.failUnless(ds.nfeatures == 50) 214 self.failUnless(ds.nsamples == 3) 215 self.failUnless((ds.labels == [1,1,1]).all()) 216 self.failUnless((ds.chunks == [0,1,2]).all()) 217 # now for the long awaited -- map back into two distinct 218 # feature spaces 219 mr = ds.mapReverse(N.arange(50)) 220 # we get two sets of feature spaces (samples and extra features) 221 self.failUnless(len(mr) == 2) 222 msamples, mxfeat = mr 223 # the sample side should be identical to the case without extra features 224 self.failUnless((msamples == N.arange(48).reshape(2,2,3,4)).all()) 225 # the extra features should be flat 226 self.failUnless((mxfeat == (48,49)).all()) 227 228 # now take a look at 229 orig = ds.O 230 self.failUnless(len(mr) == 2) 231 osamples, oextra = orig 232 self.failUnless((oextra == [[2,3],[4,5],[6,7]]).all()) 233 self.failUnless(osamples.shape == samples.shape) 234 # check that all samples not covered by an event are zero 235 filt = N.array([True,True,False,False,True,False,False,False,False,True]) 236 self.failUnless(N.sum(osamples[filt]) == 0) 237 self.failUnless((osamples[N.negative(filt)] > 0).all())

238

239 - def testEventDatasetExtended(self):

240 if not externals.exists('nifti'): 241 return 242 from mvpa.datasets.nifti import ERNiftiDataset 243 try: 244 ds = ERNiftiDataset( 245 samples=os.path.join(pymvpa_dataroot, 'bold.nii.gz'), 246 mask=os.path.join(pymvpa_dataroot, 'mask.nii.gz'), 247 events=[Event(onset=1,duration=5,label=1,chunk=1)], 248 evconv=True, tr=2.0) 249 except ValueError, e: 250 self.fail("Failed to create a simple ERNiftiDataset from a volume" 251 " with only 1 slice. Exception was:\n %s" % e)

Source Code for Module mvpa.tests.test_metadataset